You are on page 1of 741

The Conceptual Mind

The Conceptual Mind

New Directions in the Study of Concepts

edited by Eric Margolis and Stephen Laurence

The MIT Press

Cambridge, Massachusetts
London, England
© 2015 Massachusetts Institute of Technology

All rights reserved. No part of this book may be reproduced in any form by any electronic or
mechanical means (including photocopying, recording, or information storage and retrieval)
without permission in writing from the publisher.

MIT Press books may be purchased at special quantity discounts for business or sales promotional
use. For information, please email

This book was set in ITC Stone Serif Std 9/13pt by Toppan Best-set Premedia Limited, Hong Kong.
Printed and bound in the United States of America.

Library of Congress Cataloging-in-Publication Data

The conceptual mind : new directions in the study of concepts / edited by Eric Margolis and
Stephen Laurence.
pages cm
Includes bibliographical references and index.
ISBN 978-0-262-02863-9 (hardcover : alk. paper)
1. Concepts. 2. Philosophy of mind. 3. Knowledge, Theory of. I. Margolis, Eric, 1968– II. Lau-
rence, Stephen.
BD418.3.C64 2015

10 9 8 7 6 5 4 3 2 1

Contributors ix
Preface xi
Notational Convention xiii


1 Conceptual Learning by Miniature Brains 3

Aurore Avarguès-Weber and Martin Giurfa

2 Convergent Cognitive Evolution across Animal Taxa: Comparisons of

Chimpanzees, Corvids, and Elephants 29
Joshua M. Plotnik and Nicola S. Clayton

3 The Evolution of Concepts about Agents: Or, What Do Animals Recognize When
They Recognize an Individual? 57
Robert M. Seyfarth and Dorothy L. Cheney


4 Missed Connections: A Connectivity-Constrained Account of the Representation

and Organization of Object Concepts 79
Bradford Z. Mahon

5 Concept Nativism and Neural Plasticity 117

Stephen Laurence and Eric Margolis


6 The Evolution of Conceptual Design 151

H. Clark Barrett
vi Contents

7 How Natural Selection Shapes Conceptual Structure: Human Intuitions and

Concepts of Ownership 185
Pascal Boyer


8 Burge on Perception 203

Jerry A. Fodor

9 Observational Concepts 223

Daniel A. Weiskopf


10 What’s in a Concept? Analog versus Parametric Concepts in

LCCM Theory 251
Vyvyan Evans

11 Where Are the Concepts? What Words Can and Can’t Reveal 291
Barbara C. Malt, Silvia P. Gennari, Mutsumi Imai, Eef Ameel, Noburo Saji, and
Asifa Majid

12 The Representation of Events in Language and Cognition 327

Anna Papafragou


13 Relations: Language, Epistemologies, Categories, and Concepts 349

Douglas Medin, bethany ojalehto, Sandra Waxman, and Megan Bang

14 Innate Conceptual Primitives Manifested in the Languages of the World and in

Infant Cognition 379
Anna Wierzbicka


15 Why Theories of Concepts Should Not Ignore the Problem of

Acquisition 415
Susan Carey

16 Conceptual Innovation on the Frontiers of Science 455

Nancy J. Nersessian
Contents vii


17 Does the Infant Possess a Moral Concept? 477

J. Kiley Hamlin

18 Normative Concepts 519

Charles W. Kalish


19 All Concepts Are Ad Hoc Concepts 543

Daniel Casasanto and Gary Lupyan

20 By Default: Concepts Are Accessed in a Context-Independent Manner 567

Edouard Machery


21 Logical Concepts and Associative Characterizations 591

Elisabeth Camp

22 Concepts in a Probabilistic Language of Thought 623

Noah D. Goodman, Joshua B. Tenenbaum, and Tobias Gerstenberg

23 Concepts in the Semantic Triangle 655

James A. Hampton

24 Grounding Concepts 677

Frank C. Keil and Jonathan F. Kominsky

Index 693

Aurore Avarguès-Weber, Research Centre for Animal Cognition, CNRS–Université de


Eef Ameel, Faculty of Psychology and Educational Sciences, University of Leuven

Megan Bang, College of Education, University of Washington

H. Clark Barrett, Department of Anthropology and Center for Behavior, Evolution,

and Culture, UCLA

Pascal Boyer, Department of Anthropology and Department of Psychology,

Washington University in St. Louis

Elisabeth Camp, Department of Philosophy, Rutgers University

Susan Carey, Department of Psychology, Harvard University

Daniel Casasanto, Department of Psychology, University of Chicago

Nicola S. Clayton, Department of Psychology, University of Cambridge

Dorothy L. Cheney, Department of Biology, University of Pennsylvania

Vyvyan Evans, School of Linguistics & English Language, Bangor University

Jerry A. Fodor, Department of Philosophy and Rutgers Center for Cognitive Science,
Rutgers University
Silvia P. Gennari, Department of Psychology, University of York

Tobias Gerstenberg, Department of Brain and Cognitive Sciences, MIT

Martin Giurfa, Research Centre for Animal Cognition, CNRS–Université de Toulouse

Noah D. Goodman, Department of Psychology, Stanford University

J. Kiley Hamlin, Department of Psychology, University of British Columbia

James A. Hampton, Department of Psychology, City University London

x Contributors

Mutsumi Imai, Department of Environmental Information, Keio University at Shonan


Charles W. Kalish, Department of Educational Psychology, University of Wisconsin

Frank C. Keil, Department of Psychology, Yale University

Jonathan F. Kominsky, Department of Psychology, Yale University

Stephen Laurence, Department of Philosophy and Heng Centre for Cognitive Studies,
University of Sheffield

Gary Lupyan, Department of Psychology, University of Wisconsin

Edouard Machery, Department of History and Philosophy of Science, University of


Bradford Z. Mahon, Department of Brain and Cognitive Sciences, Department of

Neurosurgery, University of Rochester

Asifa Majid, Center for Language Studies and Donders Institute for Language,
Cognition, and Behaviour, Radboud University Nijmegen, and Max Planck Institute
for Psycholinguistics

Barbara C. Malt, Department of Psychology, Lehigh University

Eric Margolis, Department of Philosophy, University of British Columbia

Douglas Medin, Department of Psychology, Northwestern University

Nancy J. Nersessian, School of Interactive Computing, Georgia Institute of

Technology and Department of Psychology, Harvard University

bethany ojalehto, Department of Psychology, Northwestern University

Anna Papafragou, Department of Psychological and Brain Sciences and Department

of Linguistics and Cognitive Science, University of Delaware

Joshua M. Plotnik, Department of Psychology, University of Cambridge

Noburo Saji, Department of Child Psychology, Kamakura Women’s University

Robert M. Seyfarth, Department of Psychology, University of Pennsylvania

Joshua B. Tenenbaum, Department of Brain and Cognitive Sciences, MIT

Sandra Waxman, Department of Psychology, Northwestern University

Daniel A. Weiskopf, Department of Philosophy, Georgia State University

Anna Wierzbicka, School of Literature, Languages, and Linguistics, Australian

National University

Fifteen years have passed since the publication of Concepts: Core Readings, our previous
volume with MIT Press, and much has happened in the study of concepts since then.
The main lines of inquiry from Concepts: Core Readings have been investigated in much
greater depth, leading to many exciting findings and theoretical developments. New
lines of inquiry have also blossomed, with researchers from an ever-broader range of
disciplines making important contributions that deserve wider attention. The aim of
The Conceptual Mind: New Directions in the Study of Concepts is to take stock of these
developments and, more importantly, to look to the future.
Although The Conceptual Mind represents some of the best recent work on concepts,
it isn’t a collection of survey papers of the kind that are found in handbooks and
other reference works. Rather, our intent has been to bring together a diverse group of
leading theorists who work on concepts and have them write about an aspect of their
current views, with special emphasis on their latest work and what they take to be the
big ideas that should guide future research over the next decade. For this reason, we
view The Conceptual Mind as more of a companion to Concepts: Core Readings than a
replacement. The earlier volume includes many of the modern classics in the study
of concepts and elucidates the way in which contemporary theories emerge from an
interplay of philosophical and scientific lines of inquiry. It also identifies the major
positions that have been taken regarding the structure of concepts, and addresses the
advantages and challenges that each of these face. The Conceptual Mind, in contrast,
is devoted to the state of the art as it looks now, in 2015. Given how much the study
of concepts has opened up in recent years, it covers a more extensive range of topics
and approaches than its predecessor. For convenience, we have organized the volume
around ten major themes:

Concepts and Animals

Concepts and the Brain
Concepts and Evolution
Concepts and Perception
xii Preface

Concepts and Language

Concepts across Cultures
Concept Acquisition and Conceptual Change
Concepts and Normativity
Concepts in Context
Concepts and Conceptual Individuation

Readers who are interested in a particular area of research may use these headings
to find what they are looking for but should note that the placement of a chapter in
a given section is more a matter of emphasis than a principled distinction, as most
chapters touch on more than one of these themes.
We would like to thank all the contributors to this volume for their goodwill, their
patience, and their thought-provoking suggestions regarding where the study of con-
cepts is heading. Thanks also to Philip Laughlin and the excellent editorial team at MIT
Press for their support and for their help with this project.
Notational Convention

In this volume, small caps are used as the standard notation for specific concepts (e.g.,
ANIMAL for a concept that refers to animals).
I Concepts and Animals
1 Conceptual Learning by Miniature Brains

Aurore Avarguès-Weber and Martin Giurfa

1.1 Introduction

Concepts are considered “the glue that holds our mental life together … in that they
tie our past experiences together to our present interactions with the world” (Murphy
2002). They act as a cornerstone of human cognition and underlie analogy, language,
and mathematical abilities, among others (Lamberts and Shanks 1997). For instance,
humans and nonhuman primates learn conceptual relationships such as SAME, DIFFERENT,
LARGER THAN, BETTER THAN, and so on. In all cases, the relationships have to be encoded

by the brain independently of the physical nature of the objects linked by the relation
(Zentall, Galizio, and Critchfield 2002).
Consequently, concepts are associated with high levels of cognitive sophistica-
tion and are not, therefore, expected in an insect brain. Yet, recent works have shown
that the miniature brain of honeybees can rapidly learn conceptual relationships
between visual stimuli (Avarguès-Weber et al. 2012; Avarguès-Weber, Dyer, and Giurfa
2011; Chittka and Geiger 1995; Dacke and Srinivasan 2008; Giurfa et al. 2001; Gross
et al. 2009; Zhang et al. 2005). These results challenge the traditional view attributing
supremacy to larger brains when it comes to elaborating conceptual knowledge, and
they have, therefore, wide implications for understanding how brains can form abstract
conceptual relations in the absence of language. We address the following questions
in this chapter: (1) Which kind of evidence have we gathered for conceptual learning
in bees? (2) What do bees gain from this cognitive capacity? Is it just an epiphenom-
enon inculcated by training procedures conceived by experimenters, or is it really
adaptive and applied in an ecological context? (3) Which neurobiological mechanisms
may be underpinning this kind of problem solving in a computationally restricted bee
Before answering these questions, we start by presenting the honeybee and high-
light aspects of its behavior that have made this insect a suitable model for cognitive
4 Aurore Avarguès-Weber and Martin Giurfa

1.2 The Honeybee Apis mellifera: A Model for the Study of Simple and Higher-Order
Forms of Associative Learning

Insects have historically fascinated biologists because they offer the possibility of study-
ing sophisticated behaviors (Avarguès-Weber, Deisig, and Giurfa 2011; Chittka and
Niven 2009) and simultaneously accessing the neural bases of such behaviors (Busto,
Cervantes-Sandoval, and Davis 2010; Davis 2005; Giurfa 2007; Menzel 1999). The fact
that insects possess miniature nervous systems with a reduced number of neurons is
an advantage when the goal is the identification of networks or specific neurons that
mediate the production of behavior. Although neural circuits underlying behavior are
simple in appearance, they in fact exhibit an exquisite architecture (Strausfeld 2012).
Among insects, the honeybee has emerged as a powerful model for the study of
associative learning (Avarguès-Weber, Mota, and Giurfa 2012; Giurfa 2007; Menzel and
Giurfa 1999; Srinivasan 2010). Several reasons justify this choice, which started with
the first pioneering works of Karl von Frisch (1914). In a natural context and despite
their small size, honeybees exhibit an extremely rich behavioral repertoire (von Frisch
1967). Their lifestyle is social and relies on a complex division of labor achieved by
reproductive (queen and drones) and nonreproductive individuals (workers) (Winston
1987). Aged workers forage for food (nectar and/or pollen), which they bring back to
the colony. In these activities, a bee forager travels over distances of several kilometers
and visits hundreds of flowers in quick and efficient succession. It also collects resin
or water, or roams for information-gathering purposes. Sensory capacities and motor
performances are highly developed (Galizia, Eisenhardt, and Giurfa 2011). Bees see the
world in color (Avarguès-Weber, Mota, and Giurfa 2012; Menzel and Backhaus 1991),
discriminate shapes and patterns (Srinivasan 1994), and resolve movements with a
high temporal resolution (Srinivasan, Poteser, and Kral 1999). Their olfactory sense
is able to distinguish a large range of odors (Guerrieri et al. 2005; Vareschi 1971), and
only taste perception seems to be limited (de Brito Sanchez 2011; Robertson and Wan-
ner 2006).
In a natural context, bees learn and memorize the local cues characterizing the
places of interest, which are essentially the hive and the food sources (Menzel 1985;
Menzel, Greggers, and Hammer 1993). In the case of food sources, learning and
memory are the very basis of floral constancy, a behavior exhibited by foragers that
consists in exploiting a unique floral species as long as it offers profitable nectar or
pollen reward (Chittka, Thomsen, and Waser 1999; Grant 1951). Learning and memo-
rizing the sensory cues of the exploited flower through their association with nectar
or pollen reward is what allows a bee forager to track a particular species in the field.
When flower profitability diminishes, bees switch to a different species, thus showing
the capacity to extinguish previously learned associations.
Honeybees communicate information about important locations around the hive
through ritualized body movements, called the “waggle dance,” which transmit
Conceptual Learning by Miniature Brains 5

information about distance to and direction toward an attractive food source or nest
site (von Frisch 1967). Hive bees recruited by the waggle dance decode from the speed
of dance movement the distance to the food source and from the angle of the waggling
phase relative to gravity the flight direction relative to the sun. In this context, learning
about food sources is also possible within the hive as recruited bees learn to associate
the odor of nectar brought by a dancer with the nectar that it regurgitates and passes
them through trophallactic (mouth-to-mouth) contacts (Farina, Gruter, and Diaz 2005;
Gil and De Marco 2005).
The complexity and richness of the honeybee’s life is therefore appealing in terms of
the opportunities it offers for the study of learning and memory. Such an appeal would
be useless, however, if these phenomena would not be amenable to controlled labora-
tory conditions. Several protocols have been developed to allow experimental access in
terms of controlled training and testing conditions, thus underlining the remarkable
plasticity of this insect, which can learn even under restrictive (in terms of movement,
for instance) or stressful (in terms of the aversive reinforcement experienced) condi-
tions (Giurfa 2007; Giurfa and Sandoz 2012).
Here we will focus on visual learning protocols in which free-flying honeybees
are trained to choose visual targets paired with sucrose solution as the equivalent of
nectar reward. Experiments on concept learning in bees have all used this training

1.3 Visual Learning in Free-Flying Honeybees

Free-flying honeybees can be conditioned to visual stimuli such as colors, shapes and
patterns, depth and motion contrast, among others (Giurfa and Menzel 1997; Lehrer
1997; Srinivasan 1994; von Frisch 1914; Wehner 1981). In such a protocol, each bee
is individually marked by means of a color spot on the thorax or the abdomen so that
individual performances can be recorded. The marked bee is generally displaced by
the experimenter toward the training and test place, where it is rewarded with sucrose
solution to promote its regular return (figure 1.1). Such pretraining is performed with-
out presenting the training stimuli in order to avoid uncontrolled learning. When the
bee starts visiting the experimental place actively (i.e., without being displaced by the
experimenter), the training stimuli are presented and the choice of the appropriate
visual target reinforced with sucrose solution. Bees are trained and tested individu-
ally to achieve a precise control of the experience of each subject. It is also important
to control the distance at which a choice is made because orientation and choice
are mediated by different visual cues at different distances or angles subtended by the
target (Giurfa and Menzel 1997).
In this protocol, it is possible to gather hundreds of decisions in a relatively short
time. A motivated bee will perform dozens of visits to the feeding site in a day, return-
ing from the hive every 5 to 10 minutes, approximately. Several behaviors can be used
6 Aurore Avarguès-Weber and Martin Giurfa

Figure 1.1
An individually marked free-flying honeybee collecting sucrose solution on a concentric-disc

to quantify the bees’ choices in these experiments. Touches (i.e., the flights toward a
target that end with a contact of the bee’s antennae or legs with the stimulus surface)
and landings on a given stimulus are usually recorded to this end. The associations built
in these contexts can be operant, classical, or both, that is, they may link visual stimuli
(conditioned stimulus, CS) and reward (unconditioned stimulus, US), the response of
the animal (e.g., landing) and the US, or both. The experimental framework is never-
theless mainly operant because the bee’s behavior is the determinant for whether the
sucrose reinforcement is obtained.

1.4 Conceptual Learning in Honeybees

Recent work has shown that the miniature brain of bees can rapidly learn conceptual
relationships between visual stimuli. Concept learning is particularly interesting for
the study of a supposedly limited brain because it relies on relations between objects
Conceptual Learning by Miniature Brains 7

(Zentall, Galizio, and Critchfield 2002; Zentall et al. 2008) and requires transfer of a
rule independently of the physical nature of the stimuli considered (colors, shape, size,
etc.) (Murphy 2002, 2010). Solving conceptual problems poses, therefore, a problem for
simplistic views depicting bees, and insects in general, as rather rigid machines devoid
of plasticity.
Various recent reports have indicated that honeybees learn conceptual rules of
different sorts. These include SAMENESS/DIFFERENCE (Giurfa et al. 2001), ABOVE/BELOW
(Avarguès-Weber, Dyer, and Giurfa 2011), and the mastering of two rules simulta-
neously, ABOVE/BELOW (or RIGHT/LEFT) and DIFFERENT FROM (Avarguès-Weber et al. 2012).
Similarly, recent reports on “numerosity” in bees suggest a capacity to extract informa-
tion about number irrespective of the physical features of the objects counted (Chittka
and Geiger 1995; Dacke and Srinivasan 2008; Gross et al. 2009).

1.4.1 Sameness/Difference Concepts

The learning of the concepts of SAMENESS and DIFFERENCE was demonstrated through the
protocols of delayed matching to sample (DMTS) and delayed nonmatching to sample
(DNMTS), respectively (Giurfa et al. 2001). In these protocols an animal is presented
a nonreinforced sample and has afterwards to choose among two or more stimuli,
one of which corresponds to the sample previously shown. If trained in a DMTS, the
animal has to choose the stimulus matching the sample to obtain a positive reinforce-
ment; if trained in a DNMTS, it has to choose the opposite to the sample to obtain the
Honeybees were trained to enter a Y-maze to collect sucrose solution on one of the
arms of the maze (figure 1.2a, plate 1); the position of the reward changed randomly
between the arms of the maze from visit to visit. In a first experiment, individually
marked bees were trained following a DMTS protocol to determine if they were capable
of learning a concept of SAMENESS. Bees were presented with a changing nonrewarded
sample (i.e., one of two different color discs—“color group”—or one of two different
black-and-white gratings, vertical or horizontal—“pattern group”) at the entrance of a
maze (figure 1.2b, plate 1). The bee was rewarded only if it chose the stimulus identical
to the sample once within the maze. Bees trained with colors and presented in transfer
tests with black-and-white gratings that they had not experienced before solved the
problem and chose the grating identical to the sample at the entrance of the maze.
Similarly, bees trained with the gratings and tested with colors in transfer tests also
solved the problem and chose the novel color corresponding to that of the sample
grating at the maze entrance (figure 1.2c, plate 1). Transfer was not limited to different
types of visual stimuli (pattern vs. color), but could also operate between drastically
different sensory modalities, such as olfaction and vision (Giurfa et al. 2001). Bees
also mastered a DNMTS task, thus showing that they learn a rule of DIFFERENCE between
stimuli as well (Giurfa et al. 2001).
8 Aurore Avarguès-Weber and Martin Giurfa

Figure 1.2 (plate 1)

SAMENESS learning in honeybees (Giurfa et al. 2001). (a) Y-maze used to train bees in a delayed

matching-to-sample task. Bees had to enter the maze to collect sugar solution on one of the back
walls. A sample was shown at the maze entrance before bees accessed the arms of the maze. (b) Train-
ing protocol. A group of bees were trained during sixty trials with black-and-white vertical and
horizontal gratings (pattern group); another group was trained with colors, blue and yellow (color
group). After training, both groups were subjected to a transfer test with novel stimuli (patterns for
bees trained with colors, colors for bees trained with patterns). (c) Performance of the pattern group
and the color group in the transfer tests with novel stimuli. Both groups chose the novel stimulus
corresponding to the sample although they had no experience with such test stimuli. *: p < 0.05.
Conceptual Learning by Miniature Brains 9

These results were the first to document that bees learn rules relating stimuli in
their environment. The results were later verified in experiments showing that bees
categorize visual images based on general features common to these images (Zhang
et al. 2004) and in a study showing that the working memory underlying the solving
of the DMTS task lasts for approximately five seconds (Zhang et al. 2005), a period that
coincides with the duration of other visual and olfactory short-term memories charac-
terized in simpler forms of associative learning in honeybees (Menzel 1999).

1.4.2 Numerosity
The ability to abstract information about number has always been at the core of
animal cognition studies (Davis and Perusse 1988). The broader question underlying
this research is whether animals encode information about number when they are
presented with stimuli that are variable in other respects (Gallistel and Gelman 2000).
Several studies have shown that animals, including rats, lions, and various species
of primates, have an approximate sense of number (referred to as “numerosity”). Are
insects also endowed with this capacity?
A pioneer study on honeybee navigation suggested that bees are capable of
counting when passing successive landmarks en route to the goal and that this numer-
ical information may be used to decide when to land on a feeding place (Chittka
and Geiger 1995). Bees were trained to fly along a 300-meter transect, along which
were four identical tents equally spaced, to reach an artificial feeder placed between
the third and fourth tents. After training, the spacing between tents was increased,
thus creating a potential conflict: either the bees relied on the exact distance previ-
ously flown, in which case they should land between the second and third tents, or
they relied on the number of tents passed en route to the goal, in which case they
should land after the third tent as during training but now increasing considerably
the distance flown. Most but not all (78%; 80 of 103 bees) of the bees landed at the
correct distance; a significant percentage (22%; 23 of 103) chose to fly more than 100
meters farther and landed after passing the correct number of three tents. The intensive
training along the route may have favored attending to the real distance rather than to
the number of tents passed; yet, for some bees, the latter criterion mediated their
decision to land. The performance of these bees was taken as evidence for “proto-
counting” because it met basic criteria in most definitions of true counting, except the
fundamental transfer experiment showing that the numerical performance was inde-
pendent of the kind of objects used as landmarks to indicate the way to the goal
(Chittka and Geiger 1995). The performance of bees was not, in any event, a case of
serial recognition of different landmarks, that is, a succession of different snapshots
defining a route, because the tents passed en route to the goal were all identical; the
only way bees could decide where to land was, therefore, through a form of counting
applied to the tents.
10 Aurore Avarguès-Weber and Martin Giurfa

More recently, two new works have explored this ability in honeybees. In one of
them (Dacke and Srinivasan 2008), bees were trained to fly into a tunnel to find a food
reward after a given number of landmarks (e.g., yellow stripes on the walls and floor
of the tunnel), thus reproducing in a reduced scale the rationale of the previous tent
experiment. Yet here, the shape, size, and positions of the landmarks were changed
in the different testing conditions to avoid confounding factors. Bees showed a stron-
ger preference to land after the correct number of landmarks in nonrewarded tests
irrespective of the distance flown within the tunnel, and even in conditions in which
the landmarks were perceptually distinguishable from those used during the training.
This capacity was visible up to four landmarks passed en route to the goal but not
farther, thus indicating a limit in the bees’ counting capacity (Dacke and Srinivasan
2008). These results showed that honeybees are capable of counting in a navigational
context and that such counting operates sequentially, incrementing progressively the
number of landmarks retained.
In the other recent work, honeybees had to fly into a Y-maze and choose the stimu-
lus containing the same number of items as a sample presented at the entrance of the
maze, following a delayed matching-to-sample protocol (Gross et al. 2009). Bees were
trained to match sample images of two or three visual stimuli, and they learned to
choose a stimulus with two elements if this was the cue shown at the entrance of the
maze, or a stimulus with three elements if this was the cue present at the maze entrance
(figure 1.3a, plate 2).
In these experiments, training was performed with groups of twenty bees instead
of individuals, and bees were rewarded for correct decisions during the tests, thus
introducing undesired additional training on stimuli for which bees should remain
naïve. Although arguments were provided to justify these procedures, the con-
founding factors complicate the interpretation of the results. Yet, besides these experi-
mental caveats, results seem to clearly show a sense of numerosity in bees, as insects
correctly matched the sample of three or of two elements presented at the entrance
of the maze (Gross et al. 2009). Correct matching was independent of the physical
features of the stimuli presented in the test as choice of two versus three and could
be transferred to stimuli differing in color, shape, and configuration or orientation
(figure 1.3, plate 2). Low-level cues such as cumulated area, edge length, and illusion-
ary shape similarity formed by the elements did not mediate the bees’ choice of two
versus three. Interestingly, although bees were able to match a sample with three
elements versus an unknown stimulus presenting four elements, they were unable
to match a sample with four elements per se, even when presented against a compet-
ing stimulus with three elements. Experiments in which higher numbers were shown
also resulted in unsuccessful performances, thus revealing that three was the numer-
osity limit exhibited by the bees in the delayed matching-to-sample task (Gross et al.
Conceptual Learning by Miniature Brains 11

Figure 1.3 (plate 2)

Numerosity in honeybees (Gross et al. 2009). (a) Training protocol performed in a Y-maze (see
figure 1.2a). Bees were trained in a delayed matching-to-sample task to match stimuli containing
two or three elements. The sample with two or three elements was shown at the maze entrance
before bees accessed the arms of the maze. The bees had to choose the arm containing the stimu-
lus composed of the same number of elements as the sample to obtain sucrose reward. The appear-
ance of the elements and their spatial positions differed between the sample and the target (one
example shown) so that bees had to focus on number and not on other perceptual cues to solve
the task. (b) In transfer tests, the bees were able to match the stimuli according to the number of
their composing elements, if numbers didn’t exceed four. ***: p < 0.001.
12 Aurore Avarguès-Weber and Martin Giurfa

1.4.3 ABOVE/BELOW Concepts

For many animals that must operate in complex natural environments, spatial con-
cepts such as RIGHT, LEFT, ABOVE, and BELOW are of crucial importance to generate appro-
priate relational displacements and orientation in their environment. A recent work
studied whether honeybees learn an ABOVE/BELOW relationship between visual stimuli
and transfer it to novel stimuli that are perceptually different from those used during
the training (Avarguès-Weber, Dyer, and Giurfa 2011). Bees were trained to fly into a
Y-maze and choose visual stimuli presented above or below a horizontal bar. Training
followed a differential conditioning procedure in which one spatial relation (e.g., target
above bar) was associated with sucrose solution, while the other relation (e.g., target
below bar) was associated with quinine solution. One group of bees was rewarded on
the target above bar relation, while another group was rewarded on the target below bar
relation. After completing the training, bees were subjected to a nonrewarded transfer
test, in which a novel target stimulus (not used during the training) was presented
above or below the bar. Despite the novelty of the test situation, which preserved the
spatial relationship to the bar as the single criterion predicting the presence of sucrose
reward, bees responded appropriately: if trained for the above relationship, they chose
the novel stimulus above the bar, and if trained for the below relationship, they chose
the novel stimulus below the bar (Avarguès-Weber, Dyer, and Giurfa 2011).
Yet, because all stimuli used during the training always appeared in the same region
of the visual field (e.g., upper field for bees trained to above and lower field for bees
trained to below), an alternative interpretation could posit that instead of learning a
conceptual relationship, bees simply relied on the statistical distributions of image
differences. A series of snapshots acquired during training would determine an average
stimulus, which would be spatially distinct between the above and the below training.
Furthermore, if bees relied on a simple cue like the center of gravity of the patterns
(Ernst and Heisenberg 1999), which would be associated with reward, the problem
becomes elemental.
Both interpretations can account for the results of the experiment described above,
thus questioning a cognitive interpretation based on concept learning. Yet, they are
ruled out by a second experiment in which instead of using a salient horizontal bar
as referent, two stimuli positioned one above the other were used so that one acted
as the target and the other as the referent (figure 1.4a). Both stimuli could be well
discriminated by the bees. The target varied from trial to trial in order to promote the
extraction of an abstract ABOVE/BELOW relationship (Avarguès-Weber, Dyer, and Giurfa
2011) (figure 1.4a). As before, one group of bees was trained to select the target above
referent relationship (above group) and another group the target below referent relation-
ship (below group). Bees rapidly learned to master their respective relationship, and
when subjected to a first transfer test in which a novel stimulus was introduced as
target, they preferred the spatial relationship for which they were trained (figure 1.4b).
Figure 1.4
ABOVE/BELOW learning in honeybees (Avarguès-Weber, Dyer, and Giurfa 2011). (a) Training proto-

col. A group of bees were trained during fifty trials to fly into a Y-maze to choose black patterns on
a white background. Patterns were a variable target disposed above or below a constant referent.
Half of the bees were rewarded on the “target above referent” relation, whereas the other half was
rewarded on the “target below referent” relation. The referent pattern was either the disc or the
cross depending on the group of bees trained. In the example shown, the referent is the cross, and
the relationship rewarded during training, indicated in pink, is ABOVE (“above the cross”). After
training, bees were subjected to three types of transfer tests with novel stimuli. (b) Performance
in the transfer tests. Bees learned the concept of ABOVE/BELOW and transferred it to novel stimuli,
fulfilling the learned relationship (Transfer Test 1). Transfer tests 2 and 3 showed that neither the
spatial location of the referent on the background nor the center of gravity of stimuli was used as
a discrimination cue to resolve the task. *: p < 0.05
14 Aurore Avarguès-Weber and Martin Giurfa

A second transfer test was performed to verify that bees used the relative position of
both target and referent and not just the fact that in most cases, the referent appeared
in the upper or lower visual field of the below or the above group, respectively. In this
second transfer test, the referent was located in the middle of the background for both
the rewarded and the nonrewarded stimulus so that it could not help the bees choose
between them (figure 1.4a). In this case, bees still appropriately chose the stimulus pair
presenting the spatial relationship for which they were trained (figure 1.4b).
Even more important was a third transfer test in which only the referent was pre-
sented in the upper or the lower part of the background (figure 1.4a) to determine
whether its absolute position was an orientation cue used by the bees, instead of its
position relative to the target, which was now absent. Had the bees relied on the center
of gravity of stimuli or the statistical distribution of images, they should choose again
correctly despite the absence of the target; if, however, the relationship between target
and referent had mediated the bees’ choices, then performance should collapse. This is
exactly what happened, and choice in this test was random (figure 1.4b), thus showing
that bees did indeed learn a spatial relationship between two stimuli, a target and a
referent (Avarguès-Weber, Dyer, and Giurfa 2011).

1.4.4 Mastering Two Concepts Simultaneously

Processing several concepts simultaneously presupposes an even higher level of cogni-
tive sophistication than dealing with one concept at a time. In a recent work (Avarguès-
Weber et al. 2012), honeybees were shown to rapidly master two abstract concepts
simultaneously, one based on spatial relationships (ABOVE/BELOW and RIGHT/LEFT), and
another based on the perception of DIFFERENCE (figure 1.5, plate 3).
Bees were trained to discriminate visual stimuli composed of two images linked by
a specific spatial relationship (either ABOVE/BELOW or RIGHT/LEFT depending on bees). To
obtain the sucrose reward, the bees had to choose the appropriate spatial relationship,
irrespective of the images defining the relationship. Importantly, the two images com-
posing a stimulus were always different (figure 1.5a, plate 3). After training, subsequent
tests showed that bees learned the rule based on the spatial relation, so that they trans-
ferred their choice to novel images never seen before if these subtended the appropriate
relationship (figure 1.5b, plate 3). Moreover, they also extracted from the training the
fact that the two images were different so that they preferred the appropriate relation-
ship defined by two different images to the same relationship defined by two identical
images (figure 1.5, plate 3). Notably, if the inappropriate relationship was presented,
in one case defined by two different images and in the other by two identical images,
bees preferred the stimulus with the two different images where, at least, the rule
of difference was preserved. Finally, in a conflictive situation in which the bees had
to choose between the appropriate spatial relationship defined by identical images
and the inappropriate relationship defined by different images, the bees demonstrated
Conceptual Learning by Miniature Brains 15

no preference (figure 1.5, plate 3). These three tests showed that bees were able to
master simultaneously two different concepts: the ABOVE/BELOW, RIGHT/LEFT concept and
the DIFFERENCE concept. They assigned them the same weight in their decision-making
process so that their choice was guided by the presence of both or at least one of the
concepts. As a consequence, performance collapsed in the conflictive situation.
As in the previous study (Avarguès-Weber, Dyer, and Giurfa 2011), a series of inter-
nal within-subject controls and simulation algorithms allowed researchers to exclude
confounding low-level cues, such as the global center of gravity, the global orientation
of the stimuli, or the retinotopic similarity between the rewarded stimuli. These results
thus demonstrated that the miniature brains of bees can extract at least two different
concepts from a set of complex pictures and combine them in a rule for subsequent
choices (Avarguès-Weber et al. 2012).

1.5 An Ecological Scenario for Conceptual Learning in Bees

The problems solved by bees in conceptual problems reveal an impressive capacity to

detect relational rules between visual targets in their environment. Are these capacities
adaptive and applied in an ecological context, or are they epiphenomena inculcated by
training protocols? Although a definitive answer to this question is so far elusive, we
can discuss possible scenarios in which conceptual relationships could be extracted and
used to improve behavioral efficiency.
Among such possible scenarios, one emerges as a potential context for conceptual
learning in the case of honeybees: the navigation scenario (Chittka and Jensen 2011).
Honeybees are central-place foragers, that is, their foraging trips start and end at the
same fixed point in space: the hive. In their foraging bouts, bees may fly several kilo-
meters and follow straight or complex routes to and back from the food source (Menzel
et al. 1996; Menzel and Giurfa 2006; Menzel et al. 2012). Although bees use sky-based
information as a navigation compass, prominent landmarks and landscape informa-
tion also define routes and determine navigation strategies (Chittka, Geiger, and Kunze
1995; Collett 1996; Collett and Zeil 1998; Dacke and Srinivasan 2008). In this context,
mastering spatial relationships to build generic representations around the hive or the
food source may be particularly useful. Extracting relationships such as ABOVE/BELOW,
SAME/DIFFERENT, or TO THE RIGHT/LEFT OF may help stabilize routes in a changing environ-

ment and insulate bees against potential disorientation induced by seasonal changes
in the aspect of landmarks.
Counting could be useful in navigation tasks where the number of landmarks
encountered during a foraging trip or near the hive may contribute to efficient orienta-
tion of free-flying bees (Chittka, Geiger, and Kunze 1995; Dacke and Srinivasan 2008).
Furthermore, it could also improve foraging through evaluation of food source profit-
ability (e.g., number of flowers within a patch) or facilitate social learning (see above)
16 Aurore Avarguès-Weber and Martin Giurfa
Conceptual Learning by Miniature Brains 17

Figure 1.5 (plate 3)

Simultaneous mastering of two concepts in honeybees (Avarguès-Weber et al. 2012). Bees learned
to use two concepts simultaneously: ABOVE/BELOW (or RIGHT/LEFT) and DIFFERENCE. (a) Bees were trained
with sucrose reward in a Y-maze to choose the stimulus presenting two different patterns (Group
1) or two different colored discs (Group 2) in an above/below (or right/left) relationship depend-
ing on the group of bees. Appearances and relative position of the patterns varied from trial to
trial. (b) After a thirty training trials, bees succeeded in transferring the spatial relation rule to
unknown stimuli presenting the appropriate spatial relationship. Bees of Group 1 trained with
patterns transferred their choice to colored discs arranged in the proper spatial relationship, and
bees of Group 2 trained with colored discs transferred their choice to patterns arranged in the
proper spatial relationship. Additional tests (transfer tests 2 to 4) demonstrated that bees addition-
ally learned that the patterns linked by the spatial relation had to be different. Bees used both rules
simultaneously in transfer tests. *: p < 0.05.

by mediating foraging decisions according to the number of bees already present on a

patch of flowers (Gross et al. 2009).
This hypothesis allows predicting that not only honeybees, but also other insects
that share the capacity to navigate in a complex and structured environment and to
return constantly to the nest as a fixed point in space should be able to learn concep-
tual relationships between visual targets. This hypothesis would, nevertheless, exclude
central-place foragers that perform their foraging in an unstructured environment
(such as the Sahara Desert in the case of Cataglyphis bicolor ants; see Wehner [2003] for
review), where the possibility of relating landmarks through conceptual relationships
is rather reduced. For these insects, sky-compass-based information can provide the
essential toolkit to navigate efficiently. Similarly, central-place foragers with reduced
foraging activities (i.e., with scarce foraging bouts per unit time) are not necessarily
expected to have evolved conceptualization capacities, as their opportunities to extract
spatial relationships between objects in their environment, even if structured, would
be limited.

1.6 Neurobiological Mechanisms Underlying Conceptual Learning in Bees

Neural correlates of rules have been reported in the prefrontal cortex of both mon-
keys and rodents (Miller et al. 2003; Wallis, Anderson, and Miller 2001; White and
Wise 1999). In the dorsolateral, ventrolateral, and orbitofrontal prefrontal cortex of
monkeys, electrophysiological recordings showed the presence of neurons exhibit-
ing greater activity during “SAMENESS” trials (DMTS) as well as neurons showing greater
activity during “DIFFERENCE” trials (DNMTS), regardless of which visual sample was
used (Wallis, Anderson, and Miller 2001). In the case of insects, with no prefrontal
cortex, the critical question is which neural architecture could mediate this performance.
18 Aurore Avarguès-Weber and Martin Giurfa

While human brains have an average volume of 1,450 cm3 and comprise 100 billion
neurons, a bee brain has a volume of 1 mm3 and comprises fewer than 1 million neu-
rons. Yet higher-order associative structures allowing the combination of information
pertaining to different sensory modalities, and presenting a multimodal output con-
sistent with generalization across modalities, exist in the bee brain. Mushroom bodies
(MBs) occupy one-third of the bee brain and are candidates for mediating conceptual
learning (figure 1.6, plate 4).
In the honeybee, each MB consists of approximately 170,000 tightly packed
parallel neurons, the Kenyon cells. The bee mushroom bodies receive compartmen-
talized multisensory input (olfactory, visual, mechanosensory, gustatory) (Ehmer and
Gronenberg 2002; Mobbs 1982; Strausfeld 2002), and their extrinsic neurons, such as
the Pe1 neuron, respond to a variety of stimuli, including sucrose, odors, mechanosen-
sory stimuli, and visual stimuli (Grünewald 1999; Homberg and Erber 1979; Rybak and
Menzel 1998) (figure 1.6, plate 4). This multimodal convergence is consistent with a
capacity for integrating sensory information across various modalities and MB subcom-
partments and suits the MBs for higher-order multimodal computations, particularly
for relational associations.
Like the prefrontal cortex of primates (Barbey and Patterson 2011), MBs are
also associated with reinforcement systems and display an intrinsic relationship with
memory systems. Octopaminergic neurons, such as the VUMmx1 neuron (ventral
unpaired median neuron of the maxillary neuromere 1), which serves the function of
a reward system, converge with the regions of olfactory input of the MBs. Also, dopa-
minergic neurons, which act as a punishment system (Vergoz et al. 2007), converge

Figure 1.6 (plate 4)

Three-dimensional reconstruction of a honeybee brain. (a) The mushroom bodies, which are
multimodal structures receiving segregated visual, olfactory, mechanosensory, and gustatory
afferences, are shown in red. (b) Three-dimensional reconstruction of a mushroom body in frontal
view (vmb). Two calyces, a lateral (l) and a medial one (m), fuse in a single peduncle. Each calyx
is subdivided into three main regions, the lip, the collar (col) and the basal ring (br). (c) Scheme
of a mushroom body (delimited by dashed straight lines), showing segregated multisensory input
at the level of the calyx and integrated multimodal output at the level of the vertical (α) lobe. The
somata of the Kenyon cells (KC), which integrate the mushroom body, are located in the calyx
bowl. The dendrites of the KC form the calyx, which is subdivided into three main regions, the
lip, receiving olfactory afferences; the collar, receiving mainly visual but also mechanosensory and
gustatory afferences; and the basal ring, receiving olfactory, mechanosensory, and gustatory affer-
ences (Mobbs 1982; Schröter and Menzel 2003; Strausfeld 2002). The axons of the KC subdivide
and form the vertical (α) and the medial (β) lobe. An extrinsic, multimodal peduncle neuron, the
Pe1 neuron (Mauelshagen 1993; Okada et al. 2007; Rybak and Menzel 1998), is shown, whose
dendrites arborize across the vertical lobe; its axon projects to the lateral horn (delimited by a
dashed circle).
Conceptual Learning by Miniature Brains 19
20 Aurore Avarguès-Weber and Martin Giurfa

with specific regions of the MBs (Tedjakumala 2014). Finally, MBs have been histori-
cally characterized as a substrate for associative memories, in particular of long-term
memories (Erber, Masuhr, and Menzel 1980; Menzel, Erber, and Masuhr 1974; Menzel
and Müller 1996).
All in all, the question of whether MBs are the insect pendant of the primate prefron-
tal cortex is not relevant even if studies on the prefrontal cortex may be inspirational to
guide research on mushroom body function and vice versa. The critical question is to
determine whether and how mushroom body architecture is crucial to mediate concept
learning and which neural requisites are important to extract relational rules. From this
perspective, experiments addressing, through specific mushroom body blocking and
activation, whether or not these structures are necessary and sufficient for concept
learning, are of fundamental importance and still need to be performed.

1.7 Concept Learning in Bees: A Philosophical Perspective

We have seen that honeybees may behave as if they were guided by different kinds of
concepts inculcated by specific training procedures, that such concepts admit a plau-
sible ecological scenario, and that neural structures exist in the bee brain that could
support concept learning. From the point of view of an empirical biologist, behavior,
ecology, and neurobiology converge to favor the argument that the miniature brain
of bees, despite an apparent reduced computational power (at least in terms of the
number of neurons), is capable of elaborating conceptual knowledge.
Yet, a fundamental philosophical question one may ask is, in what sense do hon-
eybees have concepts? In other words, is the nature of concepts elaborated by bees
comparable to those available in humans? To what extent does the proposition that
we have been using all along this chapter—honeybees solve discrimination problems
using concepts—track what bees really use to master such discriminations? Is it for
instance possible that the term concept is anchored to us, humans, in a way that it is
not anchored to a bee?
The questions are important but difficult to answer because they refer to the recur-
rent problem of accessing the contents of an animal’s mind (here, an insect’s mind).
Although we may not be able to ascribe in a straightforward way a conceptual con-
tent to a bee’s mind, we can nevertheless assume that there exists a content that can
be correctly ascribed to describe efficiently the multiple discriminations reviewed in
this chapter. Because bees choose on the basis of relationships between stimuli, that
such relations may bind variable objects whose physical nature becomes irrelevant
during the problem to be solved, and that bees transfer their choice to novel situations
never seen before if the learned relationships can be detected, a content revolving
around simple discrimination learning has to be ruled out. Such discrimination learn-
ing would be stimulus specific and would preclude the kind of transfer observed in the
Conceptual Learning by Miniature Brains 21

experiments reviewed here. It thus seems safe to assume that the proposition “bees
may use concepts to solve discrimination problems” is safe and tracks what bees do,
irrespective of the nature of their conceptual knowledge.
This conclusion may create the idea that concepts are, at the end, not so elaborated
and would not represent, contrary to what we tend to believe (Murphy 2002), a higher-
order form of mental representation. This argument may be valid but has some caveats:
Do we claim that concepts are lower-form representations just because bees possess
them? In other words, if bees solve conceptual discriminations, does it have necessarily
to be on the basis of “simple” and even “primitive” representations? It is sometimes
assumed that “simple” and “miniature” nervous systems like those of insects imple-
ment cognitive faculties by radically different mechanisms compared with vertebrates,
rather relying on innate routines and elemental forms of associative learning. However,
constructing a great division between simple and advanced nervous systems will lead
us astray, because the basic logical structure of the processes underlying spontaneity,
decision making, planning, and communication are similar in many respects in big
and small brains (Chittka and Niven 2009; Menzel and Giurfa 2001).
What may be more or less demanding for the small brain of an insect is also not
obvious. For example, will it be more difficult to follow a navigation strategy based
on route following or on using a cognitive map? Is it easier to store many sequential
images defining a long route or to extract a concept connecting these images? Are
neural processes derived from behavioristic learning theory less demanding than those
derived from cognitive concepts? The answer at this stage is that we simply do not
know, and that the only way to find out is to search for neural mechanisms within a
broader conceptual frame.

1.8 Conclusion

Concept learning, described originally as a higher-order form of learning and con-

sidered a cornerstone of human cognition (Lamberts and Shanks 1997), is a capacity
that can be ascribed to honeybees based on a rich dataset spanning different forms of
conceptual problem solving. Several forms of conceptual learning have been dem-
onstrated so far in these insects, and further studies could add new forms to the list
of concepts that bees can master. The fact that only these insects have been shown
to solve learning sets leading to concept formation does not make honeybees a
cognitive exception among insects. Other insect species sharing essential traits such as
central-place navigation in structured environments, associated learning and memory
skills, and highly frequent foraging activities may perhaps be capable of comparable
In the experiments reviewed in this chapter, the performance of the bees at the
beginning of the different learning sets was always random in terms of choosing or
22 Aurore Avarguès-Weber and Martin Giurfa

not the rewarded solution versus the nonrewarded alternative. This fact underlines,
therefore, that bees did not show any preexisting bias facilitating (or impeding) the
formation of a concept. In all cases, bees learned to solve the discriminations by acquir-
ing a particular concept, thus revealing not only a high degree of behavioral plasticity,
but also that concepts are incorporated via learning. Whether bees possess, in addition,
innate concepts is difficult to determine. Innate predispositions for sensory cues such
as colors exist in bees and other pollinators and favor first encounters with flowers in
the initial foraging trips (Giurfa et al. 1995). But whether concept extraction is facili-
tated to favor efficient navigation performances remains unknown.
The case of honeybees reveals that minimal neural architectures are capable
of extracting the regularities underlying concept formation. In the absence of a pre-
frontal cortex, structures in the bee brain that are simpler in terms of their number
of neurons, but not necessarily in terms of their functional principles, can support
conceptual knowledge. The essential task is therefore to identify and characterize the
circuits that mediate concept learning in the bee brain. Such an endeavor is possible,
because reversible and targeted blocking of neural transmission in specific areas of the
bee brain is an available technical tool (Devaud et al. 2007). Coupling this procedure
with protocols inducing concept formation should enable determining the circuits that
are necessary and sufficient for achieving this task. If the honeybee has reaffirmed
its model status for learning studies through behavioral experiments demonstrating
concept learning, it can also play a significant role in unraveling the neural bases of
this capacity.


This work was supported by the Institut Universitaire de France, the French Research
Council (CNRS), and the University Paul Sabatier.


Avarguès-Weber, A., N. Deisig, and M. Giurfa. 2011. Visual cognition in social insects. Annual
Review of Entomology 56:423–443.

Avarguès-Weber, A., A. G. Dyer, M. Combe, and M. Giurfa. 2012. Simultaneous mastering of two
abstract concepts with a miniature brain. Proceedings of the National Academy of Sciences of the
United States of America 109:7481–7486.

Avarguès-Weber, A., A. G. Dyer, and M. Giurfa. 2011. Conceptualization of above and below rela-
tionships by an insect. Proceedings. Biological Sciences 278:898–905.

Avarguès-Weber, A., T. Mota, and M. Giurfa. 2012. New vistas on honey bee vision. Apidologie
Conceptual Learning by Miniature Brains 23

Barbey, A. K., and R. Patterson. 2011. Architecture of explanatory inference in the human pre-
frontal cortex. Frontiers in Psychology 2:162.

Busto, G. U., I. Cervantes-Sandoval, and R. L. Davis. 2010. Olfactory learning in Drosophila.

Physiology (Bethesda, MD) 25:338–346.

Chittka, L., and K. Geiger. 1995. Can honey bees count landmarks? Animal Behaviour

Chittka, L., K. Geiger, and J. A. N. Kunze. 1995. The influences of landmarks on distance estima-
tion of honey bees. Animal Behaviour 50:23–31.

Chittka, L., and K. Jensen. 2011. Animal cognition: Concepts from apes to bees. Current Biology

Chittka, L., and J. Niven. 2009. Are bigger brains better? Current Biology 19:R995–R1008.

Chittka, L., J. D. Thomson, and N. M. Waser. 1999. Flower constancy, insect psychology, and
plant evolution. Naturwissenschaften 86:361–377.

Collett, T. S. 1996. Insect navigation en route to the goal: Multiple strategies for the use of land-
marks. Journal of Experimental Biology 199:227–235.

Collett, T. S., and J. Zeil. 1998. Places and landmarks: An arthropod perspective. In Spatial
Representation in Animals, ed. S. Healy, 18–53. Oxford: Oxford University Press.

Dacke, M., and M. V. Srinivasan. 2008. Evidence for counting in insects. Animal Cognition

Davis, R. L. 2005. Olfactory memory formation in Drosophila: From molecular to systems neuro-
science. Annual Review of Neuroscience 28:275–302.

Davis, H., and R. Perusse. 1988. Numerical competence in animals: Definition issues, current
evidence, and a new research agenda. Behavioral and Brain Sciences 114:561–615.

de Brito Sanchez, M. G. 2011. Taste perception in honey bees. Chemical Senses 36:675–692.

Devaud, J. M., A. Blunk, J. Podufall, M. Giurfa, and B. Grünewald. 2007. Using local anaesthetics
to block neuronal activity and map specific learning tasks to the mushroom bodies of an insect
brain. European Journal of Neuroscience 26:3193–3206.

Ehmer, B., and W. Gronenberg. 2002. Segregation of visual input to the mushroom bodies in the
honeybee (Apis mellifera). Journal of Comparative Neurology 451:362–373.

Erber, J., T. Masuhr, and R. Menzel. 1980. Localization of short-term memory in the brain of the
bee, Apis mellifera. Physiological Entomology 5:343–358.

Ernst, R., and M. Heisenberg. 1999. The memory template in Drosophila pattern vision at the
flight simulator. Vision Research 39:3920–3933.

Farina, W. M., C. Gruter, and P. C. Diaz. 2005. Social learning of floral odours inside the honey-
bee hive. Proceedings. Biological Sciences 272:1923–1928.
24 Aurore Avarguès-Weber and Martin Giurfa

Galizia, C., D. Eisenhardt, and M. Giurfa. 2011. Honeybee Neurobiology and Behavior: A Tribute to
Randolf Menzel. Berlin: Springer.

Gallistel, C. R., and R. Gelman. 2000. Non-verbal numerical cognition: From the reals to integers.
Trends in Cognitive Sciences 7:307–314.

Gil, M., and R. J. De Marco. 2005. Olfactory learning by means of trophallaxis in Apis mellifera.
Journal of Experimental Biology 208:671–680.

Giurfa, M. 2007. Behavioral and neural analysis of associative learning in the honeybee: A
taste from the magic well. Journal of Comparative Physiology. A, Neuroethology, Sensory, Neural, and
Behavioral Physiology 193:801–824.

Giurfa, M., and R. Menzel. 1997. Insect visual perception: Complex abilities of simple nervous
systems. Current Opinion in Neurobiology 7:505–513.

Giurfa, M., J. A. Núñez, L. Chittka, and R. Menzel. 1995. Colour preferences of flower-naive
honeybees. Journal of Comparative Physiology. A, Neuroethology, Sensory, Neural, and Behavioral
Physiology 177:247–259.

Giurfa, M., and J. C. Sandoz. 2012. Invertebrate learning and memory: Fifty years of olfactory
conditioning of the proboscis extension response in honeybees. Learning & Memory (Cold Spring
Harbor, NY) 19:54–66.

Giurfa, M., S. Zhang, A. Jenett, R. Menzel, and M. V. Srinivasan. 2001. The concepts of
“sameness” and “difference” in an insect. Nature 410:930–933.

Grant, V. 1951. The fertilization of flowers. Scientific American 12:1–6.

Gross, H. J., M. Pahl, A. Si, H. Zhu, J. Tautz, and S. Zhang. 2009. Number-based visual generalisa-
tion in the honeybee. PLoS ONE 4:e4263.

Grünewald, B. 1999. Physiological properties and response modulations of mushroom body feed-
back neurons during olfactory learning in the honeybee Apis mellifera. Journal of Comparative
Physiology. A, Neuroethology, Sensory, Neural, and Behavioral Physiology 185:565–576.

Guerrieri, F., M. Schubert, J. C. Sandoz, and M. Giurfa. 2005. Perceptual and neural olfactory
similarity in honeybees. PLoS Biology 3:e60.

Homberg, U., and J. Erber. 1979. Response characteristics and identification of extrinsic mush-
room body neurons of the bee. Zeitschrift für Naturforschung 34:612–615.

Lamberts, K., and D. Shanks. 1997. Knowledge, Concepts, and Categories. Cambridge, MA: MIT

Lehrer, M. 1997. Honeybees’ visual spatial orientation at the feeding site. In Detection and Com-
munication in Arthropods, ed. M. Lehrer, 115–144. Basel: Birkhäuser.

Mauelshagen, J. 1993. Neural correlates of olfactory learning paradigms in an identified neuron

in the honeybee brain. Journal of Neurophysiology 69:609–625.
Conceptual Learning by Miniature Brains 25

Menzel, R. 1985. Learning in honey bees in an ecological and behavioral context. In Experimental
Behavioral Ecology and Sociobiology, ed. B. Hölldobler and M. Lindauer, 55–74. Stuttgart: Gustav
Fischer Verlag.

Menzel, R. 1999. Memory dynamics in the honeybee. Journal of Comparative Physiology. A,

Neuroethology, Sensory, Neural, and Behavioral Physiology 185:323–340.

Menzel, R., and W. Backhaus. 1991. Colour vision in insects. London: MacMillan Press.

Menzel, R., J. Erber, and T. Masuhr. 1974. Learning and memory in the honeybee. In Experimental
Analysis of Insect Behaviour, ed. L. Barton-Browne, 195–217. Berlin: Springer.

Menzel, R., K. Geiger, L. Chittka, J. Joerges, J. Kunze, and U. Muller. 1996. The knowledge base of
bee navigation. Journal of Experimental Biology 199:141–146.

Menzel, R., and M. Giurfa. 1999. Cognition by a mini brain. Nature 400:718–719.

Menzel, R., and M. Giurfa. 2001. Cognitive architecture of a mini-brain: the honeybee. Trends in
Cognitive Sciences 5:62–71.

Menzel, R., and M. Giurfa. 2006. Dimensions of cognition in an insect, the honeybee. Behavioral
and Cognitive Neuroscience Reviews 5:24–40.

Menzel, R., U. Greggers, and M. Hammer. 1993. Functional organization of appetitive learning
and memory in a generalist pollinator, the honey bee. In Insect Learning: Ecology and Evolutionary
Perspectives, ed. D. R. Papaj and A. C. Lewis, 79–125. New York: Chapman and Hall.

Menzel, R., K. Lehmann, G. Manz, J. Fuchs, M. Koblofsky, and U. Greggers. 2012. Vector integra-
tion and novel shortcutting in honeybee navigation. Apidologie 43:229–243.

Menzel, R., and U. Müller. 1996. Learning and memory in honeybees: From behavior to neural
substrates. Annual Review of Neuroscience 19:379–404.

Miller, E. K., A. Nieder, D. J. Freedman, and J. D. Wallis. 2003. Neural correlates of categories and
concepts. Current Opinion in Neurobiology 13:198–203.

Mobbs, P. G. 1982. The brain of the honeybee Apis mellifera I. The connections and spatial orga-
nization of the mushroom bodies. Philosophical Transactions of the Royal Society of London. Series B,
Biological Sciences 298:309–354.

Murphy, G. L. 2002. The Big Book of Concepts. Cambridge, MA: MIT Press.

Murphy, G. L. 2010. What are categories and concepts? In The Making of Human Concepts, ed. D.
Mareschal, P. C. Quinn, and S. E. G. Lea, 11–28. Oxford: Oxford University Press.

Okada, R., J. Rybak, G. Manz, and R. Menzel. 2007. Learning-related plasticity in PE1 and
other mushroom body-extrinsic neurons in the honeybee brain. Journal of Neuroscience 27:

Robertson, H. M., and K. W. Wanner. 2006. The chemoreceptor superfamily in the honey bee,
Apis mellifera: Expansion of the odorant, but not gustatory, receptor family. Genome Research
26 Aurore Avarguès-Weber and Martin Giurfa

Rybak, J., and R. Menzel. 1998. Integrative properties of the Pe1-neuron, a unique mushroom
body output neuron. Learning & Memory (Cold Spring Harbor, NY) 5:133–145.

Schröter, U., and R. Menzel. 2003. A new ascending sensory tract to the calyces of the honeybee
mushroom body, the subesophageal-calycal tract. Journal of Comparative Neurology 465:168–178.

Srinivasan, M. V. 1994. Pattern recognition in the honeybee: Recent progress. Journal of Insect
Physiology 40 (3): 183–194.

Srinivasan, M. V. 2010. Honey bees as a model for vision, perception, and cognition. Annual
Review of Entomology 55:267–284.

Srinivasan, M. V., M. Poteser, and K. Kral. 1999. Motion detection in insect orientation and
navigation. Vision Research 39:2749–2766.

Strausfeld, N. J. 2002. Organization of the honey bee mushroom body: Representation of the
calyx within the vertical and gamma lobes. Journal of Comparative Neurology 450:4–33.

Strausfeld, N. J. 2012. Arthropod Brains: Evolution, Functional Elegance, and Historical Significance.
Cambridge, MA: Belknap Press.

Tedjakumala, R. 2014. The role of dopamine network and signalling in aversive network in honey bee.
Ph.D. Thesis. University Toulouse 3, France.

Vareschi, E. 1971. Duftunterscheidung bei der Honigbiene—Einzelzell-Ableitungen und

Verhaltensreaktionen. Zeitschrift fur Vergleichende Physiologie 75:143–173.

Vergoz, V., E. Roussel, J. C. Sandoz, and M. Giurfa. 2007. Aversive learning in honeybees revealed
by the olfactory conditioning of the sting extension reflex. PLoS ONE 2:e288.

von Frisch, K. 1914. Der Farbensinn und Formensinn der Biene. Zoologische Jahrbucher. Abteilung
fur Allgemeine Zoologie und Physiologie der Tiere 35:1–188.

von Frisch, K. 1967. The Dance Language and Orientation of Bees. Cambridge, MA: Belknap Press.

Wallis, J. D., K. C. Anderson, and E. K. Miller. 2001. Single neurons in prefrontal cortex encode
abstract rules. Nature 411:953–956.

Wehner, R. 1981. Spatial vision in arthropods. In Handbook of Sensory Physiology VIc, ed.
H. J. Autrum, 287–616. Berlin: Springer.

Wehner, R. 2003. Desert ant navigation: How miniature brains solve complex tasks. Journal of
Comparative Physiology. A, Neuroethology, Sensory, Neural, and Behavioral Physiology 189:579–588.

White, I. M., and S. P. Wise. 1999. Rule-dependent neuronal activity in the prefrontal cortex.
Experimental Brain Research 126:315–335.

Winston, M. L. 1987. The Biology of the Honey Bee. Cambridge, MA: Harvard University Press.

Zentall, T. R., M. Galizio, and T. S. Critchfield. 2002. Categorization, concept learning, and
behavior analysis: An introduction. Journal of the Experimental Analysis of Behavior 78:237–248.
Conceptual Learning by Miniature Brains 27

Zentall, T. R., E. A. Wasserman, O. F. Lazareva, R. K. R. Thompson, and M. J. Rattermann. 2008.

Concept learning in animals. Comparative Cognition and Behavior Reviews 3:13–45.

Zhang, S., F. Bock, A. Si, J. Tautz, and M. V. Srinivasan. 2005. Visual working memory in decision
making by honey bees. Proceedings of the National Academy of Sciences of the United States of America

Zhang, S. W., M. V. Srinivasan, H. Zhu, and J. Wong. 2004. Grouping of visual objects by honey-
bees. Journal of Experimental Biology 207:3289–3298.
2 Convergent Cognitive Evolution across Animal Taxa: Comparisons
of Chimpanzees, Corvids, and Elephants

Joshua M. Plotnik and Nicola S. Clayton

2.1 Introduction

The study of intelligence has, for several decades, focused on the primate lineage. This
focus is based on an assumption that the most recent common ancestor of humans
and other nonhuman primates was subjected to environmental pressures requiring
behavioral flexibility. More recently, however, there has been increased interest in the
study of other, nonprimate groups with seemingly similar cognitive complexity. Due to
the distance between these species’ evolutionary lineages, these animals, which include
elephants (Byrne, Bates, and Moss 2009), dolphins (Marino 2002), and corvids and
parrots (Emery and Clayton 2004), must have evolved intelligence independently. So
called convergent cognitive evolution is an exciting new area of research that aims to
bridge the gap between humans and other animals in the evolution of physical and
social complexity. In general, the term “convergent cognitive evolution” implies that
certain measures of intelligence—among others, problem solving, perspective taking,
cooperation, and causal reasoning—may have evolved independently in evolutionarily
distant species that are subjected to similar selection pressures (e.g., de Waal 1996;
Emery and Clayton 2004; Marino 2002; Shettleworth 2010).
Convergent evolution explains the emergence of similar adaptations that evolve
independently in a number of different organisms under similar environmental pres-
sures (Keeton and Gould 1986). Convergence is supported by empirical evidence for
similar behavioral or anatomical traits present in a group of species, but absent in
their common ancestor. The forelimbs of birds, bats and flying insects, for instance,
all evolved into functional wings even though their common ancestor did not share
this trait (Seed, Emery, and Clayton 2009). The common ancestor of mammals and
cephalopods had simple, photoreceptive vision but not the similar, complex camera-
like eye found in its descendants (Conway Morris 2005). This suggests that convergent
evolution may shape similar, optimal adaptations (both behavioral and anatomical) in
remarkably dissimilar species. The reason for these separate yet parallel evolutionary
trajectories is most likely similar social and ecological pressures on animals living in
30 Joshua M. Plotnik and Nicola S. Clayton

similar environments. Intelligence then, for example, most likely evolved in constantly
changing environments in which an animal’s behavior would need to be able to
adapt rather quickly to change. Greater behavioral flexibility and the capacity for the
development of abstract knowledge would then not necessarily be unique to primates.
Thus, what makes the study of convergent cognitive evolution so exciting is its
departure from traditional animal models for intelligence (specifically nonhuman
primates) as well as the notion that it allows psychologists to test hypotheses about
the evolution of intelligence across unrelated animal taxa. We endeavor then to test
hypotheses on the complexity of both physical (or technical) and social cognition in
animals such as chimpanzees, corvids and elephants, for instance, by focusing on the
differences and similarities in environmental constraints to which they have been
subjected. Within the past twenty years, a growing body of empirical evidence has
emerged to support these animals’ reputation for intelligence.
In this chapter, we briefly illustrate how the field of comparative cognition is chang-
ing in light of this evidence, and use chimpanzees, elephants, and corvids as examples
to illustrate how specific animal families may have evolved similar cognitive capabili-
ties over hundreds of millions of years of evolutionary separation. This convergence
does not preclude them from developing different anatomical or behavioral traits
depending on the environments in which they lived, hence bird wings, chimpanzee
arms, and elephant trunks. These species are remarkably different in size and shape, but
here we discuss their cognitive similarities and how such analogous intelligence may
have evolved across distant taxa. Specifically, we provide selected evidence for conver-
gent cognitive evolution to illustrate the importance of continued attention to this
field, and to a more comprehensive approach to understanding how the environment
in which animals live and how specific, selective forces that have shaped flexibility in
their behavior have contributed to the evolution of their intelligence.

2.2 The How and Why of Intelligence: Physical and Social Cognition Explained?

If apes, corvids, and elephants evolved specific, complex cognitive mechanisms for
technological and social behavior, what characteristics of their environment may have
been influential in driving convergence?

2.1.1 Physical
One area of focus suggests that brain size relative to body size—often considered
a measure of intelligence (e.g., Jerison 1973)—is influenced by what, where and how
animals forage (Seed, Emery, and Clayton 2009). Primates, for instance, reliant on
fruiting plants that (a) ripen only at certain times of year, and (b) occur in only
certain, predictable areas, may have evolved intelligence specifically to locate and
remember food location and edibility (Milton 1981). Corvids are known for their
Convergent Cognitive Evolution across Animal Taxa 31

caching abilities, and their ability to remember what, where, and when they have
cached food (e.g., Clayton and Dickinson 1998, 1999) as well as who was watching
(e.g., Dally, Emery, and Clayton 2006). Elephants are known for returning to specific
food and water sources (Foley, Pettorelli, and Foley 2008; Sukumar 2003) along or
off their migration routes after periods of extreme droughts. These environmental
challenges, which must require more than basic spatiotemporal abilities, may have
shaped the evolution of intelligence.
Although finding food is one thing, actually acquiring it is another. Foraging tech-
niques and the use of tools are among the more prominent focal areas for understanding
how a need for a physical understanding of one’s environment shaped the evolution
of intelligence in animals. Byrne (1996, 2004) argued that advanced cognition has
been driven by the complexity of primate foraging and the multilevel use of tools—as
seen, for example, in gorillas (Gorilla gorilla), which must navigate plant defenses to eat
nettle leaves (Byrne 1996), and in wild chimpanzee (Pan troglodytes) nut cracking and
termite fishing (Boesch and Boesch-Achermann 2000; Whiten et al. 1999). Although
tool use is more common in the animal kingdom than once believed, this level of
complexity of tool use is reserved for only a select few species. Corvids are well known
for their tool use in the wild, and one species of crow also crafts its own tools—New
Caledonian crows (Corvus moneduloides) manufacture stepped-cut tools from leaves and
hook tools from sticks or other substrates to fish for food (Hunt 1996). In captivity,
some corvids are known to use tools even when traditionally they do not in the wild
(e.g., rooks [Corvus frugilegus]—Bird and Emery 2009a, 2009b; Eurasian jays [Garrulus
glandarius]—Cheke, Bird, and Clayton 2011).
Mendes, Hanus, and Call (2007) argued that great ape tool use is unmatched in the
animal kingdom, as corvids and other animal tool use is potentially less flexible and
complex. Seed, Emery, and Clayton (2009) argue that these species-level differences
could be need rather than capacity based. The corvid beak is, for many food-retrieval
applications, just as good as a primate-made tool or hand, and the corvid’s overall size
compared with that of the ape makes the use of some tools impractical or impossible
(Seed, Emery and Clayton, 2009). Elephants (African—Loxodonta genus, Asian—Elephas
maximus) are not well known for their tool use (but see tree branch modification for
fly swatting—Hart et al. 2001), although this is possibly because they have a built-in
tool (their trunk) that is unmatched in primates or birds. The elephant trunk has no
bones and at least forty thousand muscles, so the elephant may use its trunk in ways a
primate might use a manufactured tool. In addition, while the small body of a corvid
may prohibit it from using some tools practically, the large body size of an elephant
may make the use of some tools redundant. Elephants routinely use their body weight
to crush fruits or knock down trees to obtain food, thus making artificial tools unneces-
sary. Although tool use might not be a precursor for the evolution of complex cogni-
tion, its existence in certain species may be a measurable expression of technological
32 Joshua M. Plotnik and Nicola S. Clayton

intelligence useful for categorizing how flexibility in behavior allows animals to selec-
tively manipulate their environments.

2.1.2 Social
Animals engage in three general categories of social behavior within their environ-
ments that have most certainly shaped variability in cognition (Seed, Emery and
Clayton 2009). Competing with others to strategically maximize one’s own fitness
(“competition”—e.g., Humphrey 1976; Byrne and Whiten 1997), working together
with others for mutual gain (“cooperation”—Dugatkin 1997; de Waal 1996, 2008;
Emery et al. 2007), and learning from conspecifics (“social learning”—Galef and
Laland 2005) represent categories of behavior that require flexibility and thus may
have influenced complex cognitive development in animals. There is evidence, to
varying degrees, for all three behavioral categories in corvids (Seed, Emery, and Clayton
2009) and elephants (Poole 1996; Sukumar 2003), but here we focus on cooperation as
an example because it is the only one we have empirical, laboratory-based evidence for
in both taxa. Cooperation is relatively common in the animal kingdom, but how it is
expressed across species varies markedly (Dugatkin 1997).
Living in groups provides animals with an opportunity to increase their fitness
through cooperative efforts, an increase that would be largely unobtainable through
individual action. Cooperative hunting or predator defense is relevant here, but so is
targeted helping (assisting others in immediate need of help, for instance—de Waal
1996, 2008). The difficulty in differentiating between the cognitive pressures on
similar cooperative behavior across species—that is, how the intelligence of cooper-
ative-breeding birds or hunting lions compares to fission-fusion elephant families or
wild chimpanzee groups—makes studying them in a laboratory setting all the more
important (Seed, Emery, and Clayton 2009). Socially complex animals that may
(a) compete for food and sexual resources via alliance and coalition formation (e.g.,
Harcourt and de Waal 1992) and (b) cooperate with specific partners and sometimes
switch partners based on changes in dominance or environment make interesting test
subjects for intelligence. The dynamics of their social relationships may in fact suggest
that more complex cognitive mechanisms underlie their day-to-day decision making.
For example, socially monogamous birds that sustain long-term partnerships may end
these relationships if mating attempts fail (Mock and Fujioka 1990), suggesting they
choose to remain with partners based on the success of the cooperative partnership.
Emery and colleagues (2007) found that paired rooks regularly coordinate their behav-
ior, and Seed, Clayton, and Emery (2007) found that individuals affiliate with their
partners after the latter have been in a conflict. Seed, Emery, and Clayton (2009) argue
that this, with the result that the birds do not reconcile with other, aggressive parties,
suggests that corvids—a largely monogamous bird family—form strong social bonds
with their partners but do not form socially important relationships with others. In
Convergent Cognitive Evolution across Animal Taxa 33

addition, a recent study by Ostojić and colleagues (2013) showed that corvids select
food to share based on their partners’ current food preference and their previous meals.
The authors demonstrated that this result was due neither to the subjects’ own food
desires nor reading their partners’ behavior toward the food, suggesting the birds may
be able to attribute desire to others.
This behavior, mainly focused on a single partner, differs sharply from that of
chimpanzees, for instance, which regularly both reconcile with others they compete
with and console those with which they are affiliated (for a review, see Aureli and
de Waal 2000), an alternative social mechanism that seems to mitigate the fluc-
tuation in social relationship quality. Elephants live in largely related, matriarchal
family groups in which conflict is relatively rare (Payne 2003; Poole 1996; Plotnik and
de Waal 2014). Unlike in chimpanzee groups, changes in hierarchy are uncommon
because family groups consist of younger offspring (both male and female), and adult
females (“sisters”), led usually by one of the oldest (Douglas-Hamilton and Douglas-
Hamilton 1975; Payne 2003; Poole 1996). Thus, although social cohesion is important
in elephants (e.g., Douglas-Hamilton and Douglas-Hamilton 1975; Plotnik and de Waal
2014), the need to choose important partners (and keep track of potential competition)
may not be. Although corvids live in largely monogamous pairings and elephants live
in largely related, closely bonded family groups, this difference does not preclude either
from possessing social intelligence. Although the need for careful calculation in partner
choice may underlie the complexity of ape social intelligence, it may be less important
in monogamous corvids and female-family-centric elephants. Perhaps social complex-
ity in these species is instead driven by the physical challenges they face in their envi-
ronment, and the way in which these challenges may require coordinated efforts by or
toward partners. At the end of this chapter, we discuss some of the empirical evidence
for this idea, which suggests that although partner choice is important for determin-
ing the likelihood of success in a cooperative interaction, an ability to learn about or
understand how cooperative, joint action works may be just as relevant.
Intelligence in animals is grounded in theories on physical and social cognition, but
the study of it is reliant on controlled laboratory experiments that assess an animal’s
capacity for behavioral flexibility in a novel context. Although work on chimpanzee
cognition is decades old (for a review before 2000, see Tomasello and Call 1997; for a
discussion of more recent studies, see Seed and Tomasello 2010), a focus on conver-
gence in cognitive abilities across species is relatively new, and thus the growth in
abundance of studies on corvids and elephants is recent (Emery and Clayton 2004;
Byrne, Bates, and Moss 2009). Here, we briefly review some areas of physical and social
cognition research across chimpanzees, corvids, and elephants to show that the study
of convergent cognitive evolution is a growing field bent on understanding how com-
monalities in the environmental constraints facing evolutionarily distant species drive
similarities in intelligence.
34 Joshua M. Plotnik and Nicola S. Clayton

2.3 Comparing Cognition across Taxa: Technical Intelligence

The technical intelligence hypothesis argues that the cognitive complexity of certain
animals, specifically apes, has been shaped by the need for behavioral flexibility in
foraging techniques because of difficulties in extracting or processing food (e.g., Byrne
1997). These techniques require some sort of innovation on the part of the animal to
successfully manipulate a difficult environment.

2.3.1 Tool Use and Causal Reasoning

Until van Lawick-Goodall (1968) demonstrated that chimpanzees use physical objects
other than their own body parts to extend their normal physical reach (adapted from
the well-known definitions of tool use provided by Beck 1980; Jones and Kamil 1973),
tool use was considered a human-specific trait. Although many animal species—both
vertebrate and invertebrate—in fact use tools (Beck 1980; Emery and Clayton 2009b),
the daily use of tools for foraging is relatively rare (for reviews, see Tomasello and
Call 1997; van Horik, Clayton, and Emery 2012), and only chimpanzees, orangutans
(Pongo genus), and crows are known to use, manipulate, and manufacture tools regu-
larly in the wild (e.g., Fox, Sitompul, and Van Schaik 1999; Hunt 1996; Hunt and Gray
2004; McGrew 1992). The proximate mechanisms underlying tool use, however, are
far less well defined (Povinelli 2000). As tool-use repertoire differs among species (e.g.,
elephants using sticks to scratch or fly-switch—Chevalier-Skolnikoff and Liska 1993;
Hart et al. 2001; dolphins (Tursiops truncatus) using sponges as foraging tools—
Krützen et al. 2005; Smolker et al. 2010; or octopuses (Amphioctopus marginatus) using
self-assembled coconut-shell halves as shelter—Finn, Tregenza, and Norman 2009),
so does the cognition underlying it (Emery and Clayton 2009b; Povinelli 2000).
Chimpanzees are generally regarded as the best users of tools other than humans,1 but
there are some exciting examples of complex tool use in corvids as well (for a review,
see Emery and Clayton 2009b). New Caledonian crows are known for manipulating
leaves to make serrated probes, and both crows and rooks create and manipulate tools
specific to food-retrieval tasks (Bird and Emery 2009a, 2009b; Hunt 1996; Hunt and
Gray 2004; Weir, Chappell, and Kacelnik 2002), as well as choose specific tools based
on their size and length to solve specific problems (i.e., retrieve hidden food—Bird and
Emery 2009a; Chappell and Kacelnik 2002, 2004). The complexity of the manufacture
of such tools suggests that an animal may know its end goal before it starts to create a
tool from fresh material, but how much cognition is required to manufacture even the

1. For example, for work on cultural transmission of tool use, see Whiten et al. (1999) and
Whiten, Horner, and de Waal (2005); for variation in tool use and manufacture across wild popu-
lations (including ant dipping, nut cracking, and termite fishing), see the review in Boesch and
Tomasello (1998); and for the production of multiple tools for different purposes in the same
activity, see Sanz and Morgan (2007).
Convergent Cognitive Evolution across Animal Taxa 35

most remarkable tools in nonhuman animals is yet unknown (van Horik, Clayton, and
Emery 2012).
Many of these attempts to investigate the correlation between complex physical
cognition and tool use have occurred in the laboratory (Emery and Clayton 2009b).
Although few species regularly make tools in the wild, there is substantial innovation
(and behavioral flexibility) in other, non-tool-using animals. Rooks (Bird and Emery
2009b), Eurasian jays (Cheke, Bird, and Clayton 2011), and orangutans (Mendes, Hanus,
and Call 2007) solved an Aesop’s fable-based problem by obtaining food within a tube
using a tool. The birds, when faced with a tube containing water, used stones to raise
the water level (and thus the food), while the orangutans spit water into the tube—the
tube was dry—to raise the food to an obtainable level. Another testing paradigm, the
trap-tube task, generally requires an animal to insert a tool (or in the case of non-tool-
using animals, use a tool already part of the apparatus) to retrieve a food reward by
avoiding a trap when it is functional, or ignoring it when it is nonfunctional (Emery
and Clayton 2009b). If an animal is able to solve one version of the trap task and then
transfer its success to modifications (i.e., slight variations in trap functionality or the
relevance of specific characteristics of the traps) of the same task, this suggests rather
complex physical intelligence. For instance, Seed and colleagues (2006) showed that
rooks could solve different variations of the trap-tube task by (a) avoiding traps that
had solid bases and were thus functional, and (b) ignoring (i.e., pulling food over) traps
that had a solid top or no top or bottom at all and were thus nonfunctional. One rook
was also successful when presented with novel tube configurations, which may suggest
a more complex, causal understanding. Although rooks, crows (Taylor et al. 2009), and
chimpanzees (Seed et al. 2009) show some success on these tasks, additional studies are
needed to investigate further how well animals understand the ways in which tools
work. From a theoretical perspective, these studies suggest that the presence of tool
use in the wild is not the only predictor of overall physical intelligence. Although tool
use was once seen as a benchmark indicator of cognitive complexity, demonstrations
of physical cognition in non-tool-using species suggest that its evolution in animals
may be independent of specific behavioral traits (Emery and Clayton 2009b). Instead,
intelligent animals may demonstrate the capacity for “cognitive” and “behavioral” flex-
ibility, which allows them to solve novel and unique problems with which they other-
wise might not have prior or extensive experience (Byrne 1995; Roth and Dicke 2005).
This is an important point; as we previously discussed, tool-use demonstrations in the
wild may only occur in those species with a limited natural ability to manipulate their
physical environment. Although a recent study showed that an elephant used insight-
ful problem solving in a task requiring him to stand on a box to retrieve food (Foerder
et al. 2011), there is little evidence to date of elephants using complex tools in captivity
or the wild. But, as the technical intelligence hypothesis suggests, tools should only be
created and manipulated by those species that require them for extracting food from
their environments. Elephants are large animals with an extremely flexible built-in
36 Joshua M. Plotnik and Nicola S. Clayton

tool (their trunk), which suggests that the evolution of complex cognition in some
species may be a result of a combination of specific physical and social needs. Whereas
chimpanzee and corvid intelligence may have evolved as a result of both technical and
social environmental pressures, other species like elephants may have evolved complex
cognition more as a result of the latter.

2.4 Comparing Cognition across Taxa: Social Intelligence

Although an animal’s physical environment may affect its behavioral repertoire, and
more importantly, its cognitive range, this may not be sufficient to explain the vari-
ability of intelligence within and across species. Instead, it has been suggested that the
complexity of animals’ social relationships—specifically their need and ability to differ-
entiate between individuals, and to cooperate, deceive, and compete with others—was
an important factor in the evolution of intelligence, particularly in primates (Byrne
and Whiten 1997; Humphrey 1976). Although this “social intelligence” hypothesis,
and variations of it, has evolved over the past several decades to include a more com-
prehensive approach to understanding intelligence (e.g., including both comparative
and neuroscience approaches—see Emery et al. [2007] for a review), it has received
widespread support from those studying the evolution of intelligence in primates, and
fits well with recent evidence for complex cognition in nonprimates living in diverse
social systems.

2.4.1 Perspective Taking

The ability of animals to exploit or manipulate their social relationships is well docu-
mented in primates (de Waal 1982, 1996), but similar social complexity has also been
observed in cetaceans (Marino 2002), corvids (Emery et al. 2007), and elephants (de
Silva, Ranjeewa, and Kryazhimskiy 2011; Payne 2003; Plotnik and de Waal 2014; Poole
1996). Much of this manipulation is probably due to food competition in the wild or
a need to solve problems cooperatively as well as the ability of certain individuals to
monopolize preferred or specific food resources (Boesch and Boesch 1989; Mitani and
Watts 2001). In primate social hierarchies, there are significant fitness advantages for
individuals that are able to monopolize such food resources, in terms of access to mates
and the stability of formed alliances (Gomes and Boesch 2009; Harcourt and de Waal
1992). Thus, it makes sense that cognitive abilities would have evolved to give animals
an opportunity to manipulate their social environment to gain access to such resources,
particularly in situations where competition is especially relevant for fitness benefits
(de Waal 1982; Dugatkin 1997). Hare and colleagues (2000), for example, showed that
when subordinate and dominant chimpanzees were paired in a visual task in which
the subordinate chimpanzee could see two pieces of food while the dominant could
see only one, the subordinate individual went for the food invisible to the dominant
Convergent Cognitive Evolution across Animal Taxa 37

individual. This finding was supported by subsequent tests in which the behavior of
either the subordinate (they were released first to control for potential confounds of
dominant gaze or behavior—Hare et al. 2000), or the dominant (by replacing informed
with uninformed dominants or giving the dominant more information about the
hidden food’s location—Hare, Call, and Tomasello 2001) was manipulated. Other
examples of great ape perspective taking have focused on an ape’s ability to understand
what humans can see and know (Hare et al. 2002; Shillito et al. 2005). Together, these
results suggest that apes may recognize what others see and know and adjust their
behavior accordingly. There is the possibility, however, that potentially more parsimo-
nious explanations for the chimpanzees’ behavior exist, specifically that changes in
their behavior are due to the chimpanzees’ ability to read partner behavior rather than
an understanding of partner mental states (Emery and Clayton 2009a; Karin-D’Arcy
and Povinelli 2002; Povinelli and Vonk 2004).
Corvids are interesting models for studying perspective taking because of their
well-known food-caching strategies, which include hiding food items for future
retrieval in various locations and over varying lengths of time (de Kort and Clayton
2006; Vander Wall 1990). The flexibility in caching strategies within and across
species makes it a useful behavior for assessing complex cognition (Grodzinski and
Clayton 2010). Spatial memory underlies these birds’ ability to find caches (Clayton
1998; Shettleworth 2002) and to remember what caches have already been removed
when returning to retrieve those still present (Clayton and Dickinson 1999). The
ability to experimentally manipulate and observe caching behavior has made for
some exciting new areas of research in corvid cognition (for a review, see Emery
and Clayton 2004). From a temporal perspective (i.e., perspectives on time), corvids
have been shown to possess episodic-like memory (often referred to as the retrospec-
tive component of mental time travel) in their storage and retrieval of caches, specifi-
cally that they can remember what food they stored, when and where (Clayton and
Dickinson 1998; Clayton, Yu, and Dickinson 2001), and who was watching when they
cached (Dally, Emery, and Clayton 2006), and an ability for anticipating what they will
need in the future when deciding on their caching strategies (e.g., Correia, Dickinson,
and Clayton 2007; Raby et al. 2007; Cheke and Clayton 2012). This complex memory
for food storage gives corvids an advantage not only in temporal perspective taking but
also in social perspective taking. It has been suggested that this concept of perspective
taking may have evolved as an adaptation against cache theft (Bugnyar and Heinrich
2005; Emery and Clayton 2001)—allowing these birds to change the location and
timing of their caches based on whether or not conspecifics are present to observe.
Indeed, it has been argued that such perspective taking led to the evolution of increas-
ingly complex cache theft and cache-protection tactics via a ratchet mechanism akin to
a co-evolutionary arms race (Emery, Dally, and Clayton 2004; Shaw and Clayton 2012).
Corvids are well known for their sociality and their ability to find and pilfer conspecific
38 Joshua M. Plotnik and Nicola S. Clayton

cache stores (Bugnyar and Kotrschal 2002; Bugnyar and Heinrich 2006; Shaw and
Clayton 2012; Watanabe and Clayton 2007). The potential victims of such pilfering
have been shown to hide food in hard-to-see locations (Bugnyar and Kotrschal 2002;
Dally, Emery, and Clayton 2004, 2005), and to move caches from locations seen by pre-
vious observers (Emery and Clayton 2001, 2004, 2008). Perhaps even stronger evidence
comes from the finding that the birds only move caches if they have been observed
and, importantly, only if they have stolen another bird’s caches before (Emery and
Clayton 2001). Taken together, this body of evidence suggests the corvids’ behavior
may be a result of mental attribution or recognition of the other’s point of view (Dally,
Emery, and Clayton 2010).
Corvids may also tactically deceive other individuals by manipulating cache loca-
tions and by sham caching (going through the motions without actually hiding food—
Heinrich 1999) or changing their pilfering behavior in the presence of conspecifics
(Bugnyar and Heinrich 2005, 2006). Anecdotal and some experimental evidence for
tactical deception also exists for primates (e.g., Hare, Call, and Tomasello 2006; Whiten
and Byrne 1988), but it is surprisingly limited. In addition, emphasis on anecdotal
evidence to infer cognition is problematic (Bates and Byrne 2007; Byrne and Bates
2011; Whiten and Byrne 1988).
At this point, evidence for perspective taking in animals such as elephants and
dolphins is largely anecdotal, due to difficulties in testing these species in controlled
settings. It is important, though, to provide these examples as a starting point for future
experiments. Examples of potential perspective taking in elephants include plugging
and covering small water sources to prevent others from finding them (Shoshani 1992),
and targeted helping directed toward others in need, including lifting or coordinating
the rescue of injured or trapped conspecifics (e.g., Moss 1988, 73; Payne 1998, 64).
Similar examples of helping distressed conspecifics exist for dolphins as well (Caldwell
and Caldwell 1966, 766; Siebenaler and Caldwell 1956, 126), specifically in cases when
an incapacitated individual needs assistance reaching the water’s surface to breathe.
These observations suggest that elephants and dolphins would make interesting candi-
dates for empirical perspective-taking studies that distinguish between simple behavior
reading and the more complex, representative understanding of what a conspecific
knows, needs, or can see as a mechanism for maximizing the benefits of decision
making (de Waal 1996, 2009).
Interestingly, these species’ considerable behavioral flexibility, a common theme
throughout this chapter, would allow them to form core concepts that could be applied
to both the physical and the social world. Perspective taking and transitive inference
are two prime examples of such concepts. Transitive inference, for instance, can be
applied to the physical world (e.g., understanding that if A is bigger than B and B
is bigger than C, then A must also be bigger than C). The same conceptual knowl-
edge can also be applied to the social world in making the inference that A must be
dominant over C. Similarly, perspective taking can be applied to the physical world by
Convergent Cognitive Evolution across Animal Taxa 39

understanding that one’s view of the world changes over time, and in the social world
by understanding that another individual may have a different perspective from one’s
own. In this way, mental time travel and theory of mind can be seen as engaging the
same conceptual understanding.

2.4.2 Cue Following

Another area of research focuses on conspecific or interspecies behavioral cue follow-
ing, and the information animals use to make decisions. Theoretically, an animal’s
ability to follow conspecific cues would have important implications for decision
making based on what others see and know, and thus may have similar cognitive
underpinnings as perspective taking.
Emery and coauthors (1997) showed that rhesus monkeys (Macaca mulatta) follow
the gaze of conspecifics, and others have shown that corvids and apes both follow the
gazes of humans and move their own location to get a better look at where a human is
looking (Bräuer, Call, and Tomasello 2005; Bugnyar, Stowe, and Heinrich 2004; Schloegl,
Kotrschal, and Bugnyar 2007; see Davidson et al. 2014 for a review of gaze following
research). An often-used paradigm for assessing how animals follow social cues is the
object-choice task. A subject animal is allowed to choose one of two objects (usually
cups or buckets) under or behind which is hidden food or a target item (Call 2004; Hare
and Tomasello 1999; Miklósi and Soproni 2006). Often, these studies involve a human
experimenter who provides a limited visual cue—pointing, orienting, or gazing—to
the food or item’s hidden location. Such studies have been conducted on the great
apes (chimpanzees—Bräuer et al. 2006; Hare et al. 2002; Itakura et al. 1999; Itakura
and Tanaka 1998; Mulcahy and Call 2009; Povinelli et al. 1997; orangutans—Call and
Tomasello 1994; Itakura and Tanaka 1998; and gorillas—Peignot and Anderson 1999),
but interestingly, these studies have produced mixed results, with many suggesting
that nonhuman primates (also see monkeys—Itakura and Anderson 1996; Neiworth
et al. 2002; Vick and Anderson 2000) differ within and across species on their success
in this task. Surprisingly, many nonprimate species have been tested on similar visual
object-choice tasks and have done well by following the human-provided social cue to
find hidden food (e.g., domestic dogs, Canis familiaris—Agnetta, Hare, and Tomasello
2000; Hare and Tomasello 1999, 2005; Miklósi, Topal, and Csanyi 1998; African gray
parrots, Psittacus erithacus—Giret, Miklósi, Kreutzer, and Bovet 2009; domestic goats,
Capra hircus—Kaminski et al. 2005; horses, Equus caballus—Maros, Gacsi, and Miklósi
2008; Proops, Walton, and McComb 2010; dingoes, Canis dingo—Smith and Litchfield
2010; South African fur seals—Arctocephalus pusillus—Scheumann and Call 2004; jack-
daws, Corvus monedula—von Bayern and Emery 2009; Clark’s nutcrackers, Nucifraga
columbiana—Tornick et al. 2011; and African elephants, Loxodonta africana—Smet and
Byrne 2013, but see Plotnik et al. 2013 for negative results in Asian elephants, Elephas
maximus). Hare and colleagues (2002) showed that domesticated dogs but not hand-
reared wolves could use these same cues and argued that this discrepancy between
40 Joshua M. Plotnik and Nicola S. Clayton

species that could and could not follow human-provided social cues was most likely
due to the process of domestication, which drove the artificial selection for the ability
to follow human social cues. An animal’s ability to follow social cues in conspecifics or
humans, or the lack thereof, may not be a direct sign of intelligence but suggests that
it is important first to identify the social, behavioral cues different species use before
attempting to assess how these species employ complex cognition in decision-making

2.4.3 Cooperation
Behavioral flexibility is often an indicator of higher-level cognition that is perhaps nec-
essary for social relationships driven by changing partner roles and fluctuating levels of
cooperation and competition (de Waal 1996; Emery and Clayton 2004). The ultimate
mechanisms underlying cooperation have been well-studied (Axelrod and Hamilton
1981; Dugatkin 1997; Trivers 1971), but the proximate, cognitive mechanisms have
not. Empirical investigations of how animals “think” about cooperation generally look
at animals’ reactions to their partners’ behavior (Noë 2006). The ultimate mechanisms
underlying cooperation—for example, kin selection (altruistic behavior directed toward
genetic relatives—Hamilton 1964), byproduct mutualism (behavior, individually self-
ish, that provides immediate benefits to others—Brown 1983), and reciprocal altruism
(benefits obtained through the exchange of altruistic acts over time—Trivers 1971)—
vary across the animal kingdom; cooperation is not rare, but much of the proximate
mechanisms underlying it are most likely innately programmed and not cognitively
complex (de Waal 1996).
Crawford (1937) presented a landmark study on chimpanzees that served as the
basis for most future experimental research on cooperation in animals. Two chimpan-
zees needed to pull two ropes attached to an out-of-reach box to retrieve food placed on
top of it; the box was too heavy for a single individual’s effort, and thus a coordinated
effort by both chimpanzees was needed to pull in the box. The chimpanzees did not
immediately coordinate their pulling efforts, but after some teaching and reinforce-
ment, they learned to cooperate without human cuing. The chimpanzees nonetheless
needed to learn to coordinate their pulling on the task to successfully and consistently
obtain food. Primatologists have often debated the cognitive complexity underlying
cooperative behavior (for example, see Boesch 1994 and Stanford et al. 1994 for a
discussion of chimpanzee hunting behavior). However, in the empirical study of coop-
eration in a controlled laboratory environment, the focus is on the capacity of ani-
mals to maintain cooperative behavior without cuing from an experimenter, even if
initially, some of the behavioral mechanisms were learned during the task (Hirata and
Fuwa 2007; Melis, Hare, and Tomasello 2006). In experiments, the difference between
test and control variables often allows for the testing of whether animal performance
is due to random chance or a recognition (or understanding) of how a task works; in
cooperation, this might mean the difference between an animal cooperating by chance
Convergent Cognitive Evolution across Animal Taxa 41

or by demonstrating an understanding of partnership and coordination (Mendres and

de Waal 2000). Incidentally, this fundamental mantra of experimental psychology
is one of the most difficult aspects of designing experiments focused on comparing
cognition across species. If we cannot ask animals what they are thinking directly, we
must design controls and specific conditions within an experiment to attempt to assess
cognition empirically.
More than fifty years after Crawford (1937), Chalmeau (1994) and Chalmeau and
Gallo (1996) tested whether two chimpanzees could simultaneously pull a handle
to obtain food. Although dominance was an important factor (the dominant male
obtained most of the food), his attention to his partner during the task increased with
time, which may indicate his understanding of the need for a partner (Hirata and Fuwa
2007). Mendres and de Waal (2000) used a bar-pull apparatus (similar to Crawford’s
rope-pull task in that the bars were attached to a sliding tray) to demonstrate that
capuchin monkeys (Cebus apella) could learn about the need for partner collabora-
tion in a cooperation task, although Chalmeau, Visalberghi, and Gallo (1997) and
Visalberghi, Quarantotti, and Tranchida (2000) suggested that the monkeys only
cooperated when their random pulling occurred at the same time. Melis, Hare, and
Tomasello (2006) adopted a procedure designed by Hirata and Fuwa (2007—described
below) and found that chimpanzees not only knew they needed a partner to success-
fully retrieve food (one chimpanzee had to release another into the experiment room
by pulling a door pin, and only did so when a partner was needed), but also chose more
effective partners when a choice was available.
Hirata and Fuwa’s (2007) apparatus was a simple task similar to Crawford’s (1937)
rope pull in that two individuals needed to pull two ropes to retrieve an out-of-reach
food reward. The former’s apparatus, however, did not rely on the weight of a box
being too much for a single individual’s effort; instead, the two ropes were actually
two ends of the same rope, threaded through and around the out-of-reach food trays.
Thus,, if one rope end was pulled without the other, the entire rope would become
unthreaded and the food unobtainable. This modification made it easier to test whether
the resulting cooperation (i.e., when the food was successfully obtained by two pulling
chimpanzees) was due to random chance or to an understanding of the task. Hirata
and Fuwa (2007) first trained the chimpanzees to pull both ends together without
the help of a partner. When they were then allowed to try to pull each rope end indi-
vidually (and cooperatively) with a partner, they failed. After varying the lengths of
available rope so that shorter rope ends required faster coordination than longer ones
(because one chimpanzee’s pulling before the arrival of the partner would mean faster
failure), the chimpanzees gradually learned to coordinate their efforts. Chimpanzees
regularly looked toward their partners before pulling, and when paired with human
partners, actually recruited or solicited them for cooperation. The chimpanzees became
highly successful even when the available rope end was short and they had to wait for
their partner before pulling. This finding suggests that chimpanzees can learn how to
42 Joshua M. Plotnik and Nicola S. Clayton

cooperate in a task requiring behavioral flexibility, and to coordinate their behavior

when necessary.
Similar studies on nonprimates, although few in number thus far, have produced
varying results. In a task similar to Hirata and Fuwa (2007), rooks pulled the two rope
ends when they arrived together (Scheid and Noë 2010; Seed, Clayton, and Emery
2008). However, although there was evidence the rooks synchronized their actions in
order to solve the task, they failed to wait for a partner’s arrival if they arrived sepa-
rately, suggesting a possible lack of understanding about the need for a partner (Seed,
Clayton, and Emery 2008). In studies on hyenas (Crocuta crocuta—Drea and Carter
2009) and capuchin monkeys (Mendres and de Waal 2000) that used similar coopera-
tive pulling apparatuses, the subjects seemed to recognize that a partner’s presence was
important for obtaining food, but their successful cooperation did not necessarily indi-
cate or require an understanding that their partner’s participation in the task mattered.
Chimpanzees, however, seemed to have an understanding of both (Hare et al. 2007;
Hirata and Fuwa 2007; Melis, Hare, and Tomasello 2006). Plotnik and colleagues (2011)
tested Asian elephants and found that they pulled cooperatively and waited to pull if

Figure 2.1
Multiple viewpoints (from the ground—1; a bird’s eye view—2; from the side—3) of the elephant
cooperation apparatus, from Plotnik et al. (2011). In test and control trials, two elephants walked
down two separate roped-off lanes (represented as dashed lines in views 1 and 2). Both ends of the
same rope needed to be pulled concurrently by both elephants to successfully move the table, and
thus coordinated cooperation was needed to obtain the food rewards. Drawings by Frans B. M. de
Waal and republished with permission.
Convergent Cognitive Evolution across Animal Taxa 43

their partner arrived later (figure 2.1). In an additional condition, the elephants inhib-
ited pulling if the partner lacked rope access, and retreated when the partner failed to
pull. Various elephant pairs also exhibited different successful strategies, including one
elephant who learned that standing on the rope while the partner did all the pulling
was as successful a strategy as simultaneous coordinated pulling. Although these results
are not conclusive in predicting the cognition needed for such collaboration, elephants
perhaps also understood something about the role their partners played in the coop-
erative task. Some species, including elephants, may possess cooperative abilities that
rival those of the apes (Plotnik et al. 2011), although various aspects of proximate-level
cooperation remain to be tested.

2.5 Conclusion

2.5.1 Problems Facing Experimenters: How Animals “View” Their World

Although we have provided specific examples for chimpanzees, corvids, and elephants,
this does not preclude other animals from demonstrating convergent cognitive evolu-
tion. The obvious behavioral and physiological differences that exist in evolutionarily
distinct species, however, present an important concern for experimental psychologists—
how can one draw comparisons between species when the ecological validity of certain
paradigms differs dramatically for distant taxa? We may first look to similarities in behav-
ior across species before trying to draw psychological comparisons (Emery and Clayton
2009a, 2009b; Stevens and Hauser 2004), but even this may suggest unfair contrasts.
Some of the most classical experiments in comparative cognition, for instance,
specifically those focused on nonhuman primates, present animals with a problem to
solve. The aforementioned object-choice task in visual cue reading, the trap-tube task
in causal reasoning and tool use, and the rope-pulling task in cooperation experiments
all require the subject animal to visualize the presented problem or task in order to
solve it. If vision is necessary for even the most basic problem-solving tasks, this may
present a problem for animals that have enhanced nonvisual, or multimodal sensory
perception, including elephants (Byrne, Bates, and Moss 2009), dolphins and other
cetaceans (Marino 2002), and even pigs (Sus scrofa domesticus—Croney et al. 2003;
Held et al. 2001, 2002). Primates overwhelmingly solve problems in their environment
by processing visual information (Hare et al. 2000; Hare et al. 2001; Tomasello and
Call 1997), although primate vocal communication certainly affects social decision
making (Cheney and Seyfarth 1992; Snowdon, Brown, and Petersen 1983). The same
goes for corvids (Balda, Pepperberg, and Kamil 1998; Emery and Clayton 2004; Seed
et al. 2006), but not elephants, which may rely more heavily on olfactory information
in navigating their physical environment and acoustic information in their social world
(Bates et al. 2007; Bates et al. 2008; Langbauer 2000; McComb et al. 2000; McComb
et al. 2002; Payne 2003). Elephants present an interesting conundrum in that they are
44 Joshua M. Plotnik and Nicola S. Clayton

a highly social, cognitively complex species with a visual capacity that seems to be
inferior to their olfactory and acoustic senses (Fowler and Mikota 2006; Plotnik et al.
2014; Sukumar 2003). Although they have done well on both mirror self-recognition
(Plotnik, de Waal, and Reiss 2006) and cooperation (Plotnik et al. 2011) tasks, more
complex physical cognition paradigms may be very difficult for elephants if they
are only given sufficient visual information to investigate them (Plotnik et al. 2013).
Thus, presenting subjects with cognitive tasks that require the processing of visual
information may bias the results in favor of a negative outcome if visual information
processing—especially in natural, physical or social cognition scenarios—is ecologi-
cally invalid for the test species.
If the study of convergent cognitive evolution is to proceed across several evolu-
tionary taxa, greater attention will need to be given to the design of experiments that
can be manipulated for and then compared across species with different ecological or
sensory adaptations.

2.5.2 Final Remarks

The study of convergent cognitive evolution, or the emergence of comparable cogni-
tive complexity in evolutionarily distant species, is an exciting new area of research
that encourages collaboration among primatologists, psychologists, and biologists,
and lends important credence to the idea that humans are not alone as physically
and socially intelligent beings. In this chapter, we have briefly outlined some of
the evidence for physical and social understanding in nonhuman animals, and how
the use of behavioral flexibility allows them to apply core concepts such as perspective
taking to both the social and the physical world. This evidence strongly supports the
argument that understanding the evolution of human intelligence requires a broader
viewpoint than that drawn by traditional psychology alone. If we are to look beyond
the primate lineages to find those other nonhuman animals truly capable of being
cognitive beings, we must both assess the environmental pressures that would have
driven intelligence and find empirical ways to test hypotheses about their physical and
social cognition that allow for constructive, accurate comparisons of animal behavior.


Agnetta, B., B. Hare, and M. Tomasello. 2000. Cues to food locations that domestic dogs (Canis
familiaris) of different ages do and do not use. Animal Cognition 3:107–112.

Aureli, F., and F. B. M. de Waal, eds. 2000. Natural Conflict Resolution. Berkeley: University of
California Press.

Axelrod, R., and W. D. Hamilton. 1981. The evolution of cooperation. Science 211:1390–1396.
Convergent Cognitive Evolution across Animal Taxa 45

Balda, R. P., I. M. Pepperberg, and A. C. Kamil. 1998. Animal Cognition in Nature: The Convergence
of Psychology and Biology in Laboratory and Field. San Diego: Academic Press.

Bates, L. A., and R. W. Byrne. 2007. Creative or created: Using anecdotes to investigate animal
cognition. Methods (San Diego, Calif.) 42:12–21.

Bates, L. A., K. N. Sayialel, N. W. Njiraini, C. J. Moss, J. H. Poole, and R. W. Byrne. 2007.

Elephants classify human ethnic groups by odor and garment color. Current Biology 17 (22):

Bates, L. A., K. N. Sayialel, N. W. Njiraini, J. H. Poole, C. J. Moss, and R. W. Byrne. 2008. African
elephants have expectations about the locations of out-of-sight family members. Biology Letters 4
(1): 34–36.

Beck, B. B. 1980. Animal Tool Behaviour. New York: Garland.

Bird, C. D., and N. J. Emery. 2009a. Insightful problem solving and creative tool modification by
captive nontool-using rooks. Proceedings of the National Academy of Sciences of the United States of
America 106:10370–10375.

Bird, C. D., and N. J. Emery. 2009b. Rooks use stones to raise the water level to reach a floating
worm. Current Biology 19:1410–1414.

Boesch, C. 1994. Cooperative hunting in wild chimpanzees. Animal Behaviour 48:653–667.

Boesch, C., and H. Boesch. 1989. Hunting behavior of wild chimpanzees in the Tai National Park.
American Journal of Physical Anthropology 78:547–573.

Boesch, C., and H. Boesch-Achermann. 2000. The Chimpanzees of the Tai Forest. Oxford: Oxford
University Press.

Boesch, C., and M. Tomasello. 1998. Chimpanzee and human cultures. Current Anthropology

Bräuer, J., J. Call, and M. Tomasello. 2005. All great ape species follow gaze to distant locations
and around barriers. Journal of Comparative Psychology 119 (2): 145–154.

Bräuer, J., J. Kaminski, J. Riedel, J. Call, and M. Tomasello. 2006. Making inferences about
the location of hidden food: Social dog, causal ape. Journal of Comparative Psychology 120 (1):

Brown, J. L. 1983. Cooperation: A biologist’s dilemma. Advances in the Study of Behavior 13:

Bugnyar, T., and B. Heinrich. 2005. Ravens, Corvus corax, differentiate between knowledgeable
and ignorant competitors. Proceedings of the Royal Society, B: Biological Sciences 272:1641–1646.

Bugnyar, T., and B. Heinrich. 2006. Pilfering ravens, Corvus corax, adjust their behaviour to social
context and identity of competitors. Animal Cognition 9:369–376.
46 Joshua M. Plotnik and Nicola S. Clayton

Bugnyar, T., and K. Kotrschal. 2002. Observational learning and the raiding of food caches in
ravens (Corvus corax): Is it “tactical” deception? Animal Behaviour 64:185–195.

Bugnyar, T., M. Stowe, and B. Heinrich. 2004. Ravens, Corvus corax, follow gaze direction of
humans around obstacles. Proceedings of the Royal Society, B: Biological Sciences 271:1331–1336.

Byrne, R. W. 1995. The Thinking Ape: Evolutionary Origins of Intelligence. Oxford: Oxford University

Byrne, R. W. 1996. The misunderstood ape: Cognitive skills of the gorilla. In Reaching Into
Thought: The Minds of the Great Apes, ed. A. E. Russon, K. A. Bard, and S. T. Parker, 113–130.
Cambridge: Cambridge University Press.

Byrne, R. W. 1997. The technical intelligence hypothesis: An additional evolutionary stimulus to

intelligence? In Machiavellian Intelligence II. Extensions and Evaluations, ed. A. Whiten and R. W.
Byrne, 289–311. Cambridge: Cambridge University Press.

Byrne, R. W. 2004. The manual skills and cognition that lie behind hominid tool use. In The
Evolution of Thought: Evolutionary Origins of Great Ape Intelligence, ed. A. E. Russon and D. R. Begun,
31–44. Cambridge: Cambridge University Press.

Byrne, R. W., and L. A. Bates. 2011. Cognition in the wild: Exploring animal minds with observa-
tional evidence. Biology Letters 7 (4): 619–622.

Byrne, R. W., L. A. Bates, and C. J. Moss. 2009. Elephant cognition in primate perspective.
Comparative Cognition and Behavior Reviews 4:1–15.

Byrne, R. W., and A. Whiten. 1997. Machiavellian intelligence. In Machiavellian Intelligence II.
Extensions and Evaluations, ed. A. Whiten and R. W. Byrne, 1–23. Cambridge: Cambridge Univer-
sity Press.

Caldwell, M. C., and D. K. Caldwell. 1966. Epimeletic (care-giving) behavior in Cetacea.

In Whales, Dolphins and Porpoises, ed. K. S. Norris, 755–789. Berkeley: University of California

Call, J. 2004. Inferences about the location of food in the great apes (Pan paniscus, Pan troglodytes,
Gorilla gorilla, and Pongo pygmaeus). Journal of Comparative Psychology 118 (2): 232–241.

Call, J., and M. Tomasello. 1994. Production and Comprehension of Referential Pointing by
Orangutans (Pongo-Pygmaeus). Journal of Comparative Psychology 108 (4): 307–317.

Chalmeau, R. 1994. Do chimpanzees cooperate in a learning task? Primates 35:385–392.

Chalmeau, R., and A. Gallo. 1996. What chimpanzees (Pan troglodytes) learn in a cooperative task.
Primates 37:39–47.

Chalmeau, R., E. Visalberghi, and A. Gallo. 1997. Capuchin monkeys (Cebus apella) fail to under-
stand a cooperative task. Animal Behaviour 54:1215–1225.

Chappell, J., and A. Kacelnik. 2002. Tool selectivity in a non-primate, the New Caledonian crow
(Corvus moneduloides). Animal Cognition 5:71–78.
Convergent Cognitive Evolution across Animal Taxa 47

Chappell, J., and A. Kacelnik. 2004. Selection of tool diameter by New Caledonian crows (Corvus
moneduloides). Animal Cognition 7:121–127.

Cheke, L. G., C. D. Bird, and N. S. Clayton. 2011. Tool-use and instrumental learning in the
Eurasian jay (Garrulus glandarius). Animal Cognition 14:441–455.

Cheke, L. G., and N. S. Clayton. 2012. Eurasian jays (Garrulus glandarius) overcome their current
desires to anticipate two distinct future needs and plan for them appropriately. Biology Letters

Cheney, D. L., and R. M. Seyfarth. 1992. How Monkeys See the World: Inside the Mind of Another
Species. Chicago: University of Chicago Press.

Chevalier-Skolnikoff, S., and J. Liska. 1993. Tool use by wild and captive elephants. Animal
Behaviour 46:209–219.

Clayton, N. S. 1998. Memory and the hippocampus in food-storing birds: A comparative

approach. Neuropharmacology 37:441–452.

Clayton, N. S., and A. Dickinson. 1998. Episodic-like memory during cache recovery by scrub
jays. Nature 395:272–274.

Clayton, N. S., and A. Dickinson. 1999. Scrub jays (Aphelocoma coerulescens) remember the relative
time of caching as well as the location and content of their caches. Journal of Comparative Psychol-
ogy 113:403–416.

Clayton, N. S., K. S. Yu, and A. Dickinson. 2001. Scrub jays (Aphelocoma coerulescens) form inte-
grated memories of the multiple features of caching episodes. Journal of Experimental Psychology:
Animal Behavior Processes 27:17–29.

Conway Morris, S. 2005. Life’s Solution: Inevitable Humans in a Lonely Universe. Cambridge:
Cambridge University Press.

Correia, S. P. C., A. Dickinson, and N. S. Clayton. 2007. Western scrub-jays anticipate future
needs independently of their current motivational state. Current Biology 17:856–861.

Crawford, M. P. 1937. The cooperative solving of problems by young chimpanzees. Comparative

Psychology Monographs 14:1–88.

Croney, C. C., K. M. Adams, C. G. Washington, and W. R. Stricklin. 2003. A note on visual, olfac-
tory and spatial cue use in foraging behavior of pigs: Indirectly assessing cognitive abilities.
Applied Animal Behaviour Science 83:303–308.

Dally, J. M., N. J. Emery, and N. S. Clayton. 2004. Cache protection strategies by western scrub-
jays (Aphelocoma californica): Hiding food in the shade. Proceedings of the Royal Society, B: Biological
Sciences 271 (supplement 6): S387–S390.

Dally, J. M., N. J. Emery, and N. S. Clayton. 2005. Cache protection strategies by western
scrub-jays (Aphelocoma californica): Implications for social cognition. Animal Behaviour
48 Joshua M. Plotnik and Nicola S. Clayton

Dally, J. M., N. J. Emery, and N. S. Clayton. 2006. Food-caching western scrub-jays keep track of
who was watching when. Science 312:1662–1665.

Dally, J. M., N. J. Emery, and N. S. Clayton. 2010. Avian theory of mind and counter espionage
by food-caching western scrub-jays (Aphelocoma californica). European Journal of Developmental
Psychology 7:17–37.

Davidson, G. L., S. Butler, E. Fernández-Juricic, A. Thornton, and N. S. Clayton. 2014. Gaze sensi-
tivity: Function and mechanisms from sensory and cognitive perspectives. Animal Behaviour

de Kort, S. R., and N. S. Clayton. 2006. An evolutionary perspective on caching by corvids.

Proceedings: Biological Sciences 273:417–423.

de Silva, S., A. D. Ranjeewa, and S. Kryazhimskiy. 2011. The dynamics of social networks among
female Asian elephants. BioMed Central Ecology 11:17.

de Waal, F. B. M. 1982. Chimpanzee Politics. London: Jonathan Cape.

de Waal, F. B. M. 1996. Good Natured: The Origins of Right and Wrong in Humans and Other Animals.
Cambridge: Harvard University Press.

de Waal, F. B. M. 2008. Putting the altruism back into altruism: The evolution of empathy. Annual
Review of Psychology 59:279–300.

de Waal, F. B. M. 2009. The Age of Empathy: Nature’s Lessons for a Kinder Society. New York: Crown.

Douglas-Hamilton, I., and O. Douglas-Hamilton. 1975. Among the Elephants. New York: Viking.

Drea, C. M., and A. N. Carter. 2009. Cooperative problem solving in a social carnivore. Animal
Behaviour 78:967–977.

Dugatkin, L. A. 1997. Cooperation Among Animals: An Evolutionary Perspective. Oxford: Oxford

University Press.

Emery, N. J., and N. S. Clayton. 2001. Effects of experience and social context on prospective
caching strategies by scrub jays. Nature 414:443–446.

Emery, N. J., and N. S. Clayton. 2004. The mentality of crows: Convergent evolution of intelli-
gence in corvids and apes. Science 306:1903–1907.

Emery, N. J., and N. S. Clayton. 2008. How to build a scrub-jay that reads minds. In Origins of the
Social Mind: Evolutionary and Developmental Views, ed. S. Itakura and K. Fujita, 65–97. Tokyo:
Springer Japan.

Emery, N. J., and N. S. Clayton. 2009a. Comparative social cognition. Annual Review of Psychology

Emery, N. J., and N. S. Clayton. 2009b. Tool use and physical cognition in birds and mammals.
Current Opinion in Neurobiology 19:27–33.
Convergent Cognitive Evolution across Animal Taxa 49

Emery, N. J., J. Dally, and N. S. Clayton. 2004. Western scrub-jays (Aphelocoma californica)
use cognitive strategies to protect their caches from thieving conspecifics. Animal Cognition 7:

Emery, N. J., E. N. Lorincz, D. I. Perrett, M. W. Oram, and C. I. Baker. 1997. Gaze following
and joint attention in rhesus monkeys (Macaca mulatta). Journal of Comparative Psychology

Emery, N. J., A. M. Seed, A. M. P. von Bayern, and N. S. Clayton. 2007. Cognitive adaptations of
social bonding in birds. Philosophical Transactions of the Royal Society of London: Series B, Biological
Sciences 362:489–505.

Finn, J. K., T. Tregenza, and M. D. Norman. 2009. Defensive tool use in a coconut-carrying
octopus. Current Biology 19:R1069–R1070.

Foerder, P., M. Galloway, T. Barthel, D. E. Moore, and D. Reiss. 2011. Insightful problem solving
in an Asian elephant. PLoS ONE 6:e23251.

Foley, C., N. Pettorelli, and L. Foley. 2008. Severe drought and calf survival in elephants. Biology
Letters 4:541–544.

Fowler, M. E., and S. K. Mikota. 2006. Biology, Medicine, and Surgery of Elephants. Oxford:

Fox, E. A., A. F. Sitompul, and C. P. Van Schaik. 1999. Intelligent tool use in wild Sumatran
orangutans. In The Mentalities of Gorillas and Orangutans: Comparative Perspectives, ed. S. T. Parker,
R. W. Mitchell, and H. L. Miles, 99–116. Cambridge: Cambridge University Press.

Galef, G. G., and K. N. Laland. 2005. Social learning in animals: Empirical studies and theoretical
models. Bioscience 55:489–499.

Giret, N., A. Miklósi, M. Kreutzer, and D. Bovet. 2009. Use of experimenter-given cues by African
gray parrots (Psittacus erithacus). Animal Cognition 12 (1): 1–10.

Gomes, C. M., and C. Boesch. 2009. Wild chimpanzees exchange meat for sex on a long-term
basis. PLoS ONE 4 (4): e5116.

Grodzinski, U., and N. S. Clayton. 2010. Problems faced by food-caching corvids and the
evolution of cognitive solutions. Philosophical Transactions of the Royal Society of London: Series B,
Biological Sciences 365:977–987.

Hamilton, W. D. 1964. The genetical evolution of social behaviour I. Journal of Theoretical Biology

Harcourt, A. H., and F. B. M. de Waal. 1992. Coalitions and Alliances in Humans and Other Animals.
Oxford: Oxford University Press.

Hare, B., M. Brown, C. Williamson, and M. Tomasello. 2002. The domestication of social cogni-
tion in dogs. Science 298 (5598): 1634–1636.
50 Joshua M. Plotnik and Nicola S. Clayton

Hare, B., J. Call, B. Agnetta, and M. Tomasello. 2000. Chimpanzees know what conspecifics do
and do not see. Animal Behaviour 59:771–785.

Hare, B., J. Call, and M. Tomasello. 2001. Do chimpanzees know what conspecifics know? Animal
Behaviour 61 (1): 139–151.

Hare, B., J. Call, and M. Tomasello. 2006. Chimpanzees deceive a human competitor by hiding.
Cognition 101:495–514.

Hare, B., A. P. Melis, V. Woods, S. Hastings, and R. Wrangham. 2007. Tolerance allows bonobos
to outperform chimpanzees on a cooperative task. Current Biology 17 (7): 619–623.

Hare, B., and M. Tomasello. 1999. Domestic dogs (Canis familiaris) use human and conspecific
social cues to locate hidden food. Journal of Comparative Psychology 113 (2): 173–177.

Hare, B., and M. Tomasello. 2005. Human-like social skills in dogs? Trends in Cognitive Sciences
9 (9): 439–444.

Hart, B. L., L. A. Hart, M. McCoy, and C. R. Sarath. 2001. Cognitive behaviour in Asian elephants:
Use and modification of branches for fly switching. Animal Behaviour 62 (5): 839–847.

Heinrich, B. 1999. Mind of the Raven. New York: HarperCollins.

Held, S., M. Mendl, C. Devereux, and R. W. Byrne. 2001. Behaviour of domestic pigs in a visual
perspective taking task. Behaviour 138:1337–1354.

Held, S., M. Mendl, K. Laughlin, and R. W. Byrne. 2002. Cognition studies with pigs: Livestock
cognition and its implication for production. Journal of Animal Science 80:E10–E17.

Hirata, S., and K. Fuwa. 2007. Chimpanzees (Pan troglodytes) learn to act with other individuals in
a cooperative task. Primates 48 (1): 13–21.

Humphrey, N. K. 1976. The social function of intellect. In Growing Points in Ethology, ed.
G. Bateson and R. A. Hinde, 303–317. Cambridge, MA: Cambridge University Press.

Hunt, G. R. 1996. Manufacture and use of hook-tools by New Caledonian crows. Nature

Hunt, G. R., and R. D. Gray. 2004. The crafting of hook tools by wild New Caledonian crows.
Biology Letters 271 (S3): S88–S90.

Itakura, S., B. Agnetta, B. Hare, and M. Tomasello. 1999. Chimpanzee use of human and conspe-
cific social cues to locate hidden food. Developmental Science 2 (4): 448–456.

Itakura, S., and J. R. Anderson. 1996. Learning to use experimenter-given cues during an object-
choice task by a capuchin monkey. Cahiers De Psychologie Cognitive-Current Psychology of Cognition
15 (1): 103–112.

Itakura, S., and M. Tanaka. 1998. Use of experimenter-given cues during object-choice tasks
by chimpanzees (Pan troglodytes), an orangutan (Pongo pygmaeus), and human infants (Homo
sapiens). Journal of Comparative Psychology 112 (2): 119–126.
Convergent Cognitive Evolution across Animal Taxa 51

Jerison, H. J. 1973. Evolution of the Brain and Intelligence. New York: Academic Press.

Jones, T. B., and A. C. Kamil. 1973. Tool-making and tool-using in the northern blue jay. Science

Kaminski, J., J. Riedel, J. Call, and M. Tomasello. 2005. Domestic goats, Capra hircus, follow gaze
direction and use social cues in an object choice task. Animal Behaviour 69:11–18.

Karin-D’Arcy, M. R., and D. J. Povinelli. 2002. Do chimpanzees know what each other see? A
closer look. International Journal of Comparative Psychology 15:21–54.

Keeton, W. T., and J. L. Gould. 1986. Biological Science. 4th ed. New York: Norton.

Krützen, M., J. Mann, M. R. Heithaus, R. C. Connor, L. Bejder, and W. B. Sherwin. 2005. Cultural
transmission of tool use in bottlenose dolphins. Proceedings of the National Academy of Sciences of
the United States of America 102:8939–8943.

Langbauer, W. R. 2000. Elephant communication. Zoo Biology 19:425–445.

Marino, L. 2002. Convergence of complex cognitive abilities in cetaceans and primates. Brain,
Behavior and Evolution 59:21–32.

Maros, K., M. Gacsi, and A. Miklósi. 2008. Comprehension of human pointing gestures in horses
(Equus caballus). Animal Cognition 11 (3): 457–466.

McComb, K., C. Moss, S. Sayialel, and L. Baker. 2000. Unusually extensive networks of vocal
recognition in African elephants. Animal Behaviour 59 (6): 1103–1109.

McComb, K., D. Reby, L. Baker, C. Moss, and S. Sayialel. 2002. Long-distance communication of
acoustic cues to social identity in African elephants. Animal Behaviour 65:317–329.

McGrew, W. C. 1992. Chimpanzee Material Culture: Implications for Human Evolution. Cambridge:
Cambridge University Press.

Melis, A. P., B. Hare, and M. Tomasello. 2006. Chimpanzees recruit the best collaborators. Science

Mendes, N., D. Hanus, and J. Call. 2007. Raising the level: Orangutans use water as a tool. Biology
Letters 3 (5): 453–455.

Mendres, K. A., and F. B. M. de Waal. 2000. Capuchins do cooperate: The advantage of an intui-
tive task. Animal Behaviour 60:523–529.

Miklósi, A., and K. Soproni. 2006. A comparative analysis of animals’ understanding of the
human pointing gesture. Animal Cognition 9 (2): 81–93.

Miklósi, A., J. Topal, and V. Csanyi. 1998. Use of experimenter-given cues in dogs. Animal
Cognition 1:113–121.

Milton, K. 1981. Distribution patterns of tropical plant foods as an evolutionary stimulus to

primate mental development. American Anthropologist 83:534–548.
52 Joshua M. Plotnik and Nicola S. Clayton

Mitani, J. C., and D. P. Watts. 2001. Why do chimpanzees hunt and share meat? Animal Behaviour

Mock, D. W., and M. Fujioka. 1990. Monogamy and long-term pair bonding in vertebrates. Trends
in Ecology & Evolution 5:39–43.

Moss, C. 1988. Elephant Memories: Thirteen Years in the Life of an Elephant Family. New York:
Fawcett Columbine.

Mulcahy, N. J., and J. Call. 2009. The performance of bonobos (Pan paniscus), chimpanzees
(Pan troglodytes), and orangutans (Pongo pygmaeus) in two versions of an object-choice task.
Journal of Comparative Psychology 123 (3): 304–309.

Neiworth, J. J., M. A. Burman, B. M. Basile, and M. T. Lickteig. 2002. Use of experimenter-given

cues in visual co-orienting and in an object-choice task by a new world monkey species, cotton
top tamarins (Saguinus oedipus). Journal of Comparative Psychology 116 (1): 3–11.

Noë, R. 2006. Cooperation experiments: Coordination through communication versus acting

apart together. Animal Behaviour 71:1–18.

Ostojić, L., R. C. Shaw, L. G. Cheke, and N. S. Clayton. 2013. Evidence suggesting that desire-state
attribution may govern food sharing in Eurasian jays. Proceedings of the National Academy of
Sciences of the United States of America 110(10): 4123–4128.

Payne, K. 1998. Silent Thunder: In the Presence of Elephants. New York: Simon & Schuster.

Payne, K. 2003. Sources of social complexity in the three elephant species. In Animal Social Com-
plexity: Intelligence, Culture, and Individualized Societies, ed. F. B. M. de Waal and P. L. Tyack, 57–85.
Cambridge, MA: Harvard University Press.

Peignot, P., and J. R. Anderson. 1999. Use of experimenter-given manual and facial cues by goril-
las (Gorilla gorilla) in an object-choice task. Journal of Comparative Psychology 113 (3): 253–260.

Plotnik, J. M., and F. B. M. de Waal. 2014. Asian elephants (Elephas maximus) reassure others in
distress. PeerJ 2:e278.

Plotnik, J. M., F. B. M. de Waal, and D. Reiss. 2006. Self-recognition in an Asian elephant. Proceed-
ings of the National Academy of Sciences of the United States of America 103:17053–17057.

Plotnik, J. M., R. Lair, W. Suphachoksahakun, and F. B. M. de Waal. 2011. Elephants know when
they need a helping trunk in a cooperative task. Proceedings of the National Academy of Sciences of
the United States of America 108 (12): 5116–5121.

Plotnik, J. M., J. J. Pokorny, T. Keratimanochaya, C. Webb, H. F. Beronja, A. Hennessey, J. Hill,

et al. 2013. Visual cues given by humans are not sufficient for Asian elephants (Elephas maximus)
to find hidden food. PLoS ONE 8 (4): e61174.

Plotnik, J. M., R. C. Shaw, D. L. Brubaker, L. N. Tiller, and N. S. Clayton. 2014. Thinking with
their trunks: Elephants use smell but not sound to locate food and exclude nonrewarding alterna-
tives. Animal Behaviour 88:91–98.
Convergent Cognitive Evolution across Animal Taxa 53

Poole, J. H. 1996. Coming of Age with Elephants. New York: Hyperion.

Povinelli, D. J. 2000. Folk Physics for Apes. New York: Oxford University Press.

Povinelli, D. J., J. E. Reaux, D. T. Bierschwale, A. D. Allain, and B. B. Simon. 1997. Exploitation of

pointing as a referential gesture in young children, but not adolescent chimpanzees. Cognitive
Development 12 (4): 327–365.

Povinelli, D. J., and J. Vonk. 2004. We don’t need a microscope to explore the chimpanzees’
mind. Mind & Language 19:1–28.

Proops, L., M. Walton, and K. McComb. 2010. The use of human-given cues by domestic horses,
Equus caballus, during an object choice task. Animal Behaviour 79 (6): 1205–1209.

Raby, C. R., D. M. Alexis, A. Dickinson, and N. S. Clayton. 2007. Planning for the future by
western scrub-jays. Nature 445:919–921.

Roth, G., and U. Dicke. 2005. Evolution of the brain and intelligence. Trends in Cognitive Sciences

Sanz, C. M., and D. B. Morgan. 2007. Chimpanzee tool technology in the Goualougo Triangle,
Republic of Congo. Journal of Human Evolution 52:420–433.

Scheid, C., and R. Noë. 2010. The performance of rooks in a cooperative task depends on their
temperament. Animal Cognition 13:545–553.

Scheumann, M., and J. Call. 2004. The use of experimenter-given cues by South African fur seals
(Arctocephalus pusillus). Animal Cognition 7 (4): 224–230.

Schloegl, C., K. Kotrschal, and T. Bugnyar. 2007. Gaze following in common ravens, Corvus corax:
Ontogeny and habituation. Animal Behaviour 74:769–778.

Seed, A. M., J. Call, N. J. Emery, and N. S. Clayton. 2009. Chimpanzees solve the trap problem
when the confound of tool-use is removed. Journal of Experimental Psychology: Animal Behavior
Processes 35:23–34.

Seed, A. M., N. S. Clayton, and N. J. Emery. 2007. Third-party postconflict affiliation in rooks,
Corvus frugilegus. Current Biology 17:152–158.

Seed, A. M., N. S. Clayton, and N. J. Emery. 2008. Cooperative problem solving in rooks (Corvus
frugilegus). Proceedings of the Royal Society, B: Biological Sciences 275:1421–1429.

Seed, A. M., N. J. Emery, and N. S. Clayton. 2009. Intelligence in corvids and apes: A case of
convergent evolution? Ethology 115:401–420.

Seed, A. M., S. Tebbich, N. J. Emery, and N. S. Clayton. 2006. Investigating physical cognition in
rooks, Corvus frugilegus. Current Biology 16 (7): 697–701.

Seed, A. M., and M. Tomasello. 2010. Primate cognition. Trends in Cognitive Sciences 2:407–419.

Shaw, R. C., and N. S. Clayton. 2012. Eurasian jays (Garrulus glandarius) flexibly switch caching
and pilfering tactics in response to social context. Animal Behaviour 84:1191–1200.
54 Joshua M. Plotnik and Nicola S. Clayton

Shettleworth, S. J. 2002. Spatial behavior, food storing, and the modular mind. In The Cognitive
Animal: Empirical and Theoretical Perspectives on Animal Cognition, ed. M. Bekoff, C. Allen, and
G. M. Burghardt, 123–128. Cambridge: MIT Press.

Shettleworth, S. J. 2010. Cognition, Evolution and Behaviour. New York: Oxford University

Shillito, D. J., R. W. Shumaker, G. G. Gallup, and B. B. Beck. 2005. Understanding visual barriers:
Evidence for level 1 perspective taking in an orangutan, Pongo pygmaeus. Animal Behaviour

Shoshani, J., ed. 1992. Elephants: Majestic Creatures of the Wild. Emmaus, PA: Rodale Press.

Siebenaler, J. B., and D. K. Caldwell. 1956. Cooperation among adult dolphins. Journal of
Mammalogy 37:126–128.

Smet, A. F., and R. W. Byrne. 2013. African elephants can use human pointing cues to find
hidden food. Current Biology 23 (20): 2033–2037.

Smith, B. P., and C. A. Litchfield. 2010. Dingoes (Canis dingo) can use human social cues to locate
hidden food. Animal Cognition 13 (2): 367–376.

Smolker, R., A. Richard, R. C. Connor, J. Mann, and P. Berggren. 2010. Sponge carrying by
dolphins (Delphinidae, Tursiops sp.): A foraging specialization involving tool use? Ethology

Snowdon, C. T., C. H. Brown, and M. R. Petersen, eds. 1983. Primate Communication. Cambridge:
Cambridge University Press.

Stanford, C. B., J. Wallis, E. Mpongo, and J. Goodall. 1994. Hunting decisions in wild chimpan-
zees. Behaviour 131:1–20.

Stevens, J. R., and M. D. Hauser. 2004. Why be nice? Psychological constraints on the evolution
of cooperation. Trends in Cognitive Sciences 8 (2): 60–65.

Sukumar, R. 2003. The Living Elephants: Evolutionary Ecology, Behaviour, and Conservation. Oxford:
Oxford University Press.

Taylor, A. H., G. R. Hunt, F. S. Medina, and R. D. Gray. 2009. Do New Caledonian crows solve
physical problems through causal reasoning? Proceedings of the Royal Society, B: Biological Sciences

Tomasello, M., and J. Call. 1997. Primate Cognition. Oxford: Oxford University Press.

Tornick, J. K., B. M. Gibson, D. Kispert, and M. Wilkinson. 2011. Clark’s nutcrackers (Nucifraga
columbiana) use gestures to identify the location of hidden food. Animal Cognition 14 (1):

Trivers, R. L. 1971. The evolution of reciprocal altruism. Quarterly Review of Biology 46:35–57.
Convergent Cognitive Evolution across Animal Taxa 55

van Horik, J., N. S. Clayton, and N. J. Emery. 2012. Convergent evolution of cognition in corvids,
apes and other animals. In The Oxford Handbook of Comparative Evolutionary Psychology, ed.
J. Vonk and T. K. Shackelford, 80–101. Oxford: Oxford University Press.

van Lawick-Goodall, J. 1968. The behaviour of free-living chimpanzees in the Gombe Stream
Reserve. Animal Behaviour Monographs 1:161–311.

Vander Wall, S. B. 1990. Food Hoarding in Animals. Chicago: University of Chicago Press.

Vick, S. J., and J. R. Anderson. 2000. Learning and limits of use of eye gaze by capuchin monkeys
(Cebus apella) in an object-choice task. Journal of Comparative Psychology 114 (2): 200–207.

Visalberghi, E., B. P. Quarantotti, and F. Tranchida. 2000. Solving a cooperation task without
taking into account the partner’s behavior: The case of capuchin monkeys (Cebus apella). Journal
of Comparative Psychology 114:297–301.

von Bayern, A. M. P., and N. J. Emery. 2009. Jackdaws respond to human attentional states and
communicative cues in different contexts. Current Biology 19 (7): 602–606.

Watanabe, S., and N. S. Clayton. 2007. Observational visuospatial encoding of the cache
locations of others by western scrub-jays (Aphelocoma californica). Journal of Ethology 25:271–279.

Weir, A. A. S., J. Chappell, and A. Kacelnik. 2002. Shaping of hooks in New Caledonian crows.
Science 297:981.

Whiten, A., and R. W. Byrne. 1988. Tactical deception in primates. Behavioral and Brain Sciences

Whiten, A., J. Goodall, W. C. McGrew, T. Nishida, V. Reynolds, Y. Sugiyama, and C. Boesch.

1999. Cultures in chimpanzees. Nature 399:682–685.

Whiten, A., V. Horner, and F. B. M. de Waal. 2005. Conformity to cultural norms of tool use in
chimpanzees. Nature 437:737–740.
3 The Evolution of Concepts about Agents: Or, What Do Animals
Recognize When They Recognize an Individual?

Robert M. Seyfarth and Dorothy L. Cheney

3.1 Introduction

We now know that the mind of a human infant is neither a blank slate nor, in
William James’s words, “one great blooming, buzzing confusion” (1890, 462). Instead,
infants are born with what Carey (2010, 2011) has called a set of “innate, representa-
tional primitives” (2011, 113) that guide infants’ learning and their expectations of
how objects are likely to behave. Three core systems of knowledge have been suggested,
each specialized for representing and reasoning about entities of different kinds (Carey
and Spelke 1996). One core system deals with the causal and spatial relations among
objects, another concerns number, and a third deals with agents. Here, we focus on
core systems of knowledge about agents—their goals, attentional states, and the causal
mechanisms that underlie their behavior.
The prevalence in human infants of a core system of knowledge about agents
suggests that at some point in our evolutionary history, a predisposition to recognize
certain entities in the world as capable of self-generated motion, with goals and the
motivation to achieve them (Carey 2011; Frith and Frith 2012), gave some individu-
als a fitness advantage over others. In modern humans, the advantages to be gained
from this style of thinking are obvious. How might it have benefited our nonhuman,
prelinguistic ancestors?
Old World monkeys live in societies where survival and reproduction depend on
social skills. To succeed, an individual must be able to predict the behavior of others,
and to do this she must understand their social relationships. Here we propose that the
demands of social life have favored individuals who can recognize other animals’ social
attributes and treat these properties as inextricable parts of an individual’s identity. To
illustrate our argument, we focus on female baboons and their recognition of other
individuals’ dominance rank and kinship.
Rank and kinship, we suggest, are examples of social concepts—baboons’ represen-
tational primitives. Recognizing these attributes in others, however, poses a problem
because there are no overt, physical markers of a female’s social position. Instead,
58 Robert M. Seyfarth and Dorothy L. Cheney

animals acquire this knowledge by observing other individuals’ social interactions and
learning the statistical regularities associated with them. Knowledge of other animals’
rank and kinship, among their other attributes, permits baboons to predict how those
animals are likely to behave when interacting with other group members. We can see
in the social cognition of baboons the evolutionary origins of our theory of mind, our
concepts about agents, and our inclination to classify other individuals in part accord-
ing to both their social relationships and the intentions we attribute to them.

3.2 Individual Recognition

Individual recognition is widespread in animals (Tibbetts and Dale 2007). Many

species have specialized brain cells that respond particularly strongly to faces (Tsao
et al. 2006; Leopold and Rhodes 2010), voices (Petkov et al. 2008), and familiar speakers
(Belin and Zattore 2003). Although such recognition has most often been documented
in the auditory mode through playback experiments (e.g., Cheney and Seyfarth 1980;
Rendall, Rodman, and Emond 1996), subjects in these experiments often seem to be
engaged in more complex cross-modal or even multimodal processing. A baboon who
looks toward the source of the sound when she hears her offspring’s call (Cheney and
Seyfarth 2007) acts as if the sound has created an expectation of what she will see if she
looks in that direction.
The first evidence that animals might integrate multiple cues to form a representa-
tion of an individual came from work by Johnston and Bullock (2001) on hamsters.
Golden hamsters have at least five different odors that are individually distinctive. In
a typical experiment, a male familiar with females A and B was exposed (and became
habituated to) the vaginal secretions of female A. He was then tested with either A’s
or B’s flank secretions. Males tested with A’s flank secretions showed little response
(across-odor habituation); however, males tested with B’s flank secretions responded
strongly. The authors concluded that “when a male was habituated to one odor he
was also becoming habituated to the integrated representation of that individual”
(Johnston and Peng 2008, 122) and was therefore not surprised to encounter a different
odor from the same animal. Hamsters, they suggested, have an integrated, multiodor
memory of other individuals. Recent experiments indicate that direct physical contact
with an individual—not just exposure to its odors—is necessary for such memories to
develop (Johnston and Peng 2008).
But what about the representation of individuals across sensory modalities? Dogs
(Adachi, Kuwahata, and Fujita 2007) and squirrel monkeys (Adachi and Fujita 2007)
associate the faces and voices of their caretakers, rhesus macaques spontaneously
match the faces and voices of familiar conspecifics and familiar humans (Adachi and
Hampton 2011; Sliwa et al. 2011; Ghazanfar et al. 2005), and both horses and crows
associate the vocalizations of a familiar group member with the sight of that individual
The Evolution of Concepts about Agents 59

(Proops, McComb, and Reby 2009; Kondo, Izawa, and Watanabe 2012). Humans, of
course, routinely integrate the perception of faces and voices to form the rich, multi-
modal concept of a person (Campanella and Belin 2007).
Supporting these behavioral observations, neurophysiological data reveal exten-
sive connections between auditory and visual areas in mammalian brains (Cappe and
Barone 2005), particularly between those areas involved in the recognition of voices
and faces among rhesus macaques (Ghazanfar and Logothetis 2003; Ghazanfar et al.
2005; Sliwa et al. 2011) and humans (von Kriegstein et al. 2005; Blank, Anwander, and
von Kreigstein 2011). These links between face- and voice-recognition areas provide
further evidence, in both monkeys and humans, for a “cross-modal, cognitive represen-
tation” of individual identity (Sliwa et al. 2011, 1735).

3.3 Other Social Classifications

Many animals not only recognize individuals but also classify them into groups,
organizing them according to their close social bonds, linear dominance ranks, and
transient sexual relations. Baboons (Papio hamadryas spp.) provide some good examples.
Baboons live throughout Africa in groups of 50 to 150 individuals. Males and females
have very different life histories. Males emigrate to other groups at around eight to
ten years of age. Females, in contrast, remain in their natal group throughout their
lives, maintaining close bonds with their matrilineal kin through frequent grooming,
mutual support in coalitions, tolerance at feeding sites, and interactions with each
other’s infants (Cheney and Seyfarth 2007; Silk et al. 2010a). Adult females can also
be ranked in a stable, linear dominance hierarchy that determines priority of access to
resources. From birth, daughters acquire ranks similar to those of their mothers. As a
result, the stable core of a baboon group consists of a hierarchy of matrilines, in which
all members of, say, matriline B outrank or are outranked by all members of matrilines
C and A, respectively. Rank relations are generally stable over time, with few reversals
occurring either within or between families. When reversals do occur, however, their
consequences differ significantly depending on who is involved. For example, if the
third-ranking female in matriline B (B3) rises in rank above her second-ranking sister
(B2), the reversal affects only these individuals; the B family’s rank relative to other
families remains unchanged. However, a rank reversal between females from different
matrilines (for example, C1 rising in rank above B3) usually causes all members of matri-
line C to rise above all members of matriline B (Cheney and Seyfarth 1990, 2007). The
ranked, matrilineal society of baboons is typical of many Old World monkeys.
Among baboon females, the ability to form stable, enduring social bonds can increase
individuals’ reproductive success. At two long-term study sites in Kenya and Botswana,
females had highly differentiated relationships with other females in their group. Some
pairs—usually but not always matrilineal kin—interacted often, maintaining close
60 Robert M. Seyfarth and Dorothy L. Cheney

bonds that lasted for many years, while others interacted infrequently. Those with
the most stable, enduring relationships had higher offspring survival (Silk, Alberts, and
Altmann 2003, Silk et al. 2009) and lived longer (Silk et al. 2010b) than females with
weaker relationships. High dominance rank also had a significant effect on longevity
but was a less powerful predictor than relationship quality.
Baboons, then, are born into a social world that is filled with statistical regularities:
animals interact in highly predictable ways. Natural selection has favored individu-
als who can recognize these patterns, because knowing about other animals’ relation-
ships is both the best way to predict their behavior and essential to forming the kind
of stable, enduring bonds that lead to high reproductive success. Field experiments
have shown that, by the time she is an adult, a female baboon recognizes the close
bonds among matrilineal kin (Cheney and Seyfarth 1999; see also below) and the
linear rank relations both among females (Cheney, Seyfarth, and Silk 1995) and among
males (Kitchen, Cheney, and Seyfarth 2005). When females hear the vocalizations
of a juvenile involved in an aggressive interaction, they respond by looking at the juve-
nile’s mother (Cheney and Seyfarth 1999; see also Cheney and Seyfarth 1980). Simi-
larly, when baboons hear a sequence of aggressive and submissive calls that mimic a
higher-ranking animal threatening a lower-ranking animal, they respond only briefly.
If the calls of the participants are reversed, however, subjects respond significantly
more strongly, as if the apparent rank reversal violates their expectations (Cheney,
Seyfarth, and Silk 1995; a control sequence ruled out the possibility that the violating
sequence evoked a stronger response simply because it was rare).
To test whether subjects classify females simultaneously according to both matrilin-
eal kinship and dominance rank, Bergman and colleagues (2003) played sequences of
calls mimicking within- and between-matriline rank reversals to subjects in matched
trials. In one trial, subjects heard an apparent rank reversal involving two members of
the same matriline: for example, female B3 giving threat-grunts and female B2 scream-
ing. In another trial, the same subject heard an apparent rank reversal involving the
members of two different matrilines: for example, female C1 giving threat-grunts and
female B3 screaming. As a control, subjects heard a fight sequence that was consistent
with the female dominance hierarchy. To control for the rank distance separating the
subject and the individual whose calls were being played, each subject heard a rank
reversal (either within or between family) that involved the matriline one step above
her own (cf. Penn, Holyoak, and Povinelli 2008). Within this constraint, the rank
distance separating apparent opponents within and between families was systemati-
cally varied.
Listeners responded with apparent surprise to sequences of calls that appeared to
violate the existing dominance hierarchy. Moreover, between-family rank reversals
elicited a consistently stronger response than did within-family rank reversals (Bergman
The Evolution of Concepts about Agents 61

et al. 2003). Subjects acted as if they classified individuals simultaneously according to

both kinship and rank. The classification of individuals simultaneously according to
these two criteria has also been documented in Japanese macaques (Schino, Tiddi, and
Polizzi di Sorrentino 2006).
These results are difficult to explain without assuming that, when one baboon
hears another vocalize, the listener encodes information not just about the caller’s
identity but also about her dominance rank and family membership, among many
other attributes. This encoding is immediate and occurs automatically: just as we
cannot hear a word without thinking about its meaning, a baboon cannot hear a
vocalization without thinking about the animal who is calling, what she looks like,
and her rank and family membership. These features are an inextricable part of the
caller’s identity, bound together in much the same way that auditory and visual cues
are bound together in a cross-modal cognitive percept. Individual recognition thus
constitutes a form of object perception (e.g., Bregman 1990; Miller and Cohen 2010), in
which a variety of disparate stimuli are linked together to form a coherent object. As a
result, perception of one of the object’s attributes (for example, her voice) creates a rich
variety of expectations in the perceiver’s mind of—for instance—what she will see
when she looks toward the sound, whom the caller is likely to dominate, and who
is likely to support her in an aggressive interaction. Individual recognition, then, is
more than just the recognition of an individual. It includes the recognition of that
individual’s place in society.

3.4 Social Concepts

What mechanisms underlie animals’ knowledge of the relations that exist among
others? One hypothesis argues that memory and classical conditioning are entirely suf-
ficient to explain primates’ social knowledge. As they mature, baboons observe behav-
iors that link individuals in predictable ways. These associations, stored in memory,
allow an observer to predict how others are likely to interact. According to this view,
the baboons’ knowledge should not be described as conceptual because, in contrast to
the case among humans, we have no independent evidence for the existence of such
concepts. Baboons’ social knowledge is therefore best explained by relatively simple
hypotheses based on learned associations and prodigious memory (e.g., Schusterman
and Kastak 1998).
Explanations based on memory and associative learning are powerful and appeal-
ing under simplified laboratory conditions, but they strain credulity when applied to
behavior in nature, where animals confront more complex sets of stimuli. A young
baboon, for example, must learn thousands of dyadic (and tens of thousands of
triadic) relations in order to predict other animals’ behavior. The magnitude of the
62 Robert M. Seyfarth and Dorothy L. Cheney

problem makes one wonder whether simple associations, even coupled with pro-
digious memory, are equal to the task (Seyfarth and Cheney 2001). Faced with the
problem of memorizing a huge, ever-changing dataset, humans (Mandler 1967) and
rats (Macuda and Roberts 1995) are predisposed to search for a higher-order rule that
makes the task easier. Why should other animals be any different?
In fact, several results suggest that, even if it begins with relatively simple Pavlov-
ian associations, primates’ social knowledge is rapidly organized into units of thought
that resemble our concepts (Dasser 1988). Consider, for example, the speed of animals’
reactions to events. When a baboon hears a sequence of vocalizations that violates the
dominance hierarchy, she responds within seconds (Cheney and Seyfarth 2007). When
a macaque or capuchin monkey involved in a fight tries to recruit an ally, she seems to
know almost immediately which individuals would be the most effective partners (Silk
1999; Perry, Barrett, and Manson 2004; Schino, Tiddi, and Polizzi di Sorrentino 2006).
The speed of these reactions suggests that animals are not searching through a mas-
sive, unstructured database of simple, dyadic associations but have instead organized
their knowledge about individuals into categories, including what we call dominance
hierarchies and matrilineal (family) groups.
These categories share many features with human concepts. For example, they
cannot be reduced to any one, or even a few, perceptual attributes. High-ranking females
are not older or larger than low-ranking females, nor do they live in larger kin groups.
Males change dominance ranks often. Family members do not always look alike, sound
alike, or share any other physical features that make them easy to tell apart. None of
this variation, however, affects other animals’ classifications: a three-legged member of
family B is still a member of family B.
Nor is the classification of individuals into family groups based on different types
or rates of interaction. The members of high-ranking families are not necessarily more
aggressive than others, nor do they feed in different areas, forage together, or groom
or play more often. In fact, grooming within families can be highly variable (Silk et al.
2010a), yet this has no effect on other animals’ perception of who belongs in which
Social categories, moreover, persist despite changes in their composition. Among
baboons, for instance, the recognition of a linear, transitive hierarchy persists despite
demographic changes in the individuals who occupy each rank. Linear, transitive rank
orders and matrilineal kin groups thus qualify as concepts because, in the mind of a
baboon, their existence is independent of the individuals that compose them.
Finally, the classification of individuals is a cognitive operation that affects behavior.
When a listener hears vocalizations from two individuals interacting elsewhere, her
response depends not just on the animals’ identities but also on their ranks and family
membership (Bergman et al. 2003). Such data support the view that social concepts are
units of thought with causal power: they determine how individuals behave.
The Evolution of Concepts about Agents 63

3.5 Concepts, Expectations, and the Attribution of Motives to Others

If the formation of social concepts is adaptive, however, individuals confront a problem

because, as already noted, the entities that make up these concepts are heterogeneous.
We propose that, faced with this dilemma, natural selection has favored those individ-
uals who analyze social interactions according to causal relations between behaviors,
and who categorize others at least in part according to their perceived intentions. Here
are some experiments that lead us to these conclusions.

3.5.1 Rank Reversal and the Violation of Expectation

The rank reversal experiments described above (Cheney, Seyfarth, and Silk 1995;
Bergman et al. 2003; Kitchen, Cheney, and Seyfarth 2005) all relied on the violation-
of-expectation method: the listener responded more strongly to “D2 threatens and B1
screams” than to “B1 threatens and D2 screams” because the former sequence violated
the listener’s expectations about how these individuals ought to behave toward each
other. But this logic holds only if the listener assumes both that B1’s scream was caused
by D2’s threat-grunt, and that D2’s threat-grunt indicates an aggressive intent toward
B1. Without this assumption of causality, there would be no violation of expectation.
Rank reversal experiments also suggest that listeners attribute intentions and motives
toward others: D2 has aggressive intentions toward B1, and this attribution of intent,
combined with knowledge of D2’s and B1’s relative ranks, causes the strong response. By
contrast, an alternative, simpler explanation makes no reference to a theory of mind:
D2’s threat-grunts simply indicate impending or probable aggressive behavior toward
B1. In sections 3.5.2–3.5.4, we describe experiments designed to test between these two
hypotheses. In section 3.5.5, we review some of the relevant neurophysiological data.

3.5.2 Judging the “Directedness” of a Vocalization

Primates are constantly required to make judgments about other animals’ intentions.
This demand is particularly striking in the context of vocal communication, when
listeners must make inferences about the intended recipient of another animal’s calls.
Primate groups are noisy, tumultuous societies, and an individual could not manage
her social interactions if she assumed that every vocalization she heard was directed at
her. Of course, listeners can often draw inferences about the intended target of a vocal-
ization from the direction of the caller’s gaze; however, such cues are not always avail-
able. Even in the absence of visual signals, monkeys are able to make such inferences
based on their knowledge of a signaler’s identity and the nature of recent interactions.
In one study, for example, subjects heard an aggressive threat-grunt from an indi-
vidual shortly after they had either exchanged aggression or groomed with that indi-
vidual. Subjects who heard a female’s threat-grunt shortly after grooming with her
ignored the call: that is, they acted as if they assumed that the female was threatening
64 Robert M. Seyfarth and Dorothy L. Cheney

another individual. But subjects who heard the same call after receiving aggression
responded strongly: they acted as if they assumed that the call was directed at them
and signaled further aggression (Engh et al. 2006). This result could not be explained by
a simple contingency judgment, since the prior event—the vocalization—was the same
in each case. Nor could results be explained by assuming that any prior interaction
with individual X “primed” subjects to expect further interaction with X, because prior
aggression and prior grooming affected the subjects’ responses to the vocalization in
different ways. Finally, the effects of prior behavior were specific to the subject’s former
partner: hearing the partner’s threat-grunt did not affect the subject’s behavior toward
other, previously uninvolved individuals. The simplest explanation would seem to be
that female baboons make inferences about the target of a vocalization even in the
absence of visual cues, and that the nature of prior interactions creates an expectation
on the part of the subject—an expectation that is based on the attribution of intentions
to another. After a fight, the subject assumes that her rival has aggressive intentions
toward her; after grooming, she draws the opposite conclusion (Cheney and Seyfarth

3.5.3 Judging the Intent to Reconcile

Tests of reconciliatory grunting in baboons provide further suggestion that listeners’
responses to vocalizations depend not only on the identity of the caller but also on their
assessment of the caller’s motives and the intended target of the call. In many species
of primates, aggressors will occasionally “reconcile” with their victims by extending a
friendly gesture toward them shortly after the fight. Among baboons, reconciliation
most commonly occurs in the form of a grunt. Grunts are signals of benign intent;
they lower the probability of subsequent aggression and facilitate friendly interactions
(Cheney, Seyfarth, and Silk 1995; Silk, Cheney, and Seyfarth 1996). In one playback
experiment, a female baboon that had recently been threatened heard within minutes
either the grunt of her aggressor or the grunt of another dominant female unrelated to
her aggressor. After hearing her former aggressor’s grunt, the female was more likely to
approach her aggressor and to tolerate her aggressor’s approach than after hearing the
grunt of the other, uninvolved dominant female. She acted as if she attributed friendly
motives to the aggressor, and therefore treated the call as a reconciliatory signal that
renewed aggression was unlikely (Cheney and Seyfarth 1997).
In a subsequent experiment, victims were played the grunt of a close relative of their
aggressor. In this case, too, they treated the grunt as a signal of reconciliation, respond-
ing as they would have if the aggressor herself had grunted: they were more likely to
approach the aggressor or the reconciling relative and more likely to tolerate either
individual’s approach. By contrast, females who heard the grunt of a dominant female
unrelated to the original aggressor showed no such response. Here again, subjects acted
as if they assumed that the grunt by the aggressor’s relative was directed at them as a
The Evolution of Concepts about Agents 65

signal of benign intent, and they accepted this grunt as a proxy of reconciliation with
their opponent. In other words, they acted as if they attributed some kind of shared
intention to the aggressor and her relative—one that they did not attribute to the unre-
lated female who vocalized in the control condition (Wittig, Crockford, Wikberg, et al.
2007). The alternative, behaviorist explanation based solely on learned contingencies
seems increasingly unlikely. If, after receiving aggression from B1, subjects respond as
if they expect further aggression from B1 (Engh et al. 2006), but after receiving aggres-
sion from B1 and hearing a grunt from B2, subjects respond as if they expect no further
aggression from B1 (Wittig, Crockford, Wikberg, et al. 2007), the difference cannot be
based on B1’s prior reconciliatory behavior because B1 has not reconciled. The differ-
ence can only be explained if we assume that subjects have different expectations—or
ascribe different motives—to B1.

3.5.4 Judging the Intention of Alliance Partners

Finally, in a test of vocal alliances among baboons, a subject who had recently been
threatened by a more dominant female heard either the aggressive threat-grunt of a
close relative of her opponent or the threat-grunt of a female belonging to a different
matriline. Subjects responded more strongly in the first condition, avoiding both the
signaler, the original antagonist, and other members of her family for a significantly
longer time than in the control condition (Wittig, Crockford, Seyfarth, et al. 2007).
Once again, subjects changed their behavior toward another individual based on an
interaction not with the individual herself but with one of the individual’s close kin.
Subjects acted as if they attributed some kind of shared intention to closely related
individuals—a motivation that they did not attribute to others who belonged to dif-
ferent matrilines.

3.5.5 The Attribution of Motives

When deciding “Who, me?” upon hearing a vocalization, baboons must take into
account the identity of the signaler (who is it?), the type of call given (friendly or
aggressive?), the nature of their prior interactions with the signaler (were they aggres-
sive, friendly, or neutral?), and the correlation between past interactions and future
ones (does a recent grooming interaction lower or increase the likelihood of aggres-
sion?). Learned contingencies doubtless play a role in these assessments. But because
listeners’ responses depend on simultaneous consideration of all these factors, it seems
likely that these assessments are also based at least in part on inferences about other
individuals’ motives and intentions.
Neurobiological research supports this hypothesis. In both monkeys and humans,
the perception of gaze direction and goal-directed behavior appear to activate the same
areas of the brain, including the superior temporal sulcus (STS) and the amygdala. The
STS is particularly sensitive to the orientation of another individual’s eyes (Jellema
66 Robert M. Seyfarth and Dorothy L. Cheney

et al. 2000; Emery and Perrett 2000). Mutual gaze evokes greater activity in the STS than
does averted gaze, suggesting that the STS facilitates the processing of social informa-
tion (Pelphrey, Viola, and McCarthy 2004). In both monkeys and humans, STS also
responds to goal-directed actions and perceptions. Cells in monkeys’ STS show par-
ticularly increased activity to goal-directed hand movement when the actor they are
observing is gazing at his or her hand (Jellema et al. 2000; Lorincz et al. 2005). It there-
fore seems possible that STS may be involved in representing what others see and what
their actions and intentions are (Gallagher and Frith 2003). Similarly, in both monkeys
and humans. the amygdala responds strongly to social stimuli, particularly aversive
ones. It also seems to be important for processing information about gaze direction
(Adolphs, Russell, and Tranel 1999; Kawashima et al. 1999; Fine, Lumsden, and Blair
2001; Santos, Flombaum, and Phillips 2006).
Other areas of monkeys’ brains seem to be sensitive to the intentions that underlie
behavior. As in humans, mirror neurons in the inferior parietal lobule (IPL) of mon-
keys’ brains are activated both when a monkey performs a specific action and when
it observes someone else performing that action. Furthermore, neurons that code for
specific acts, such as grasping, seem to be context dependent. Some mirror neurons in
monkeys respond more when they grasp a piece of food to eat it than when they grasp
the same food to place it into a container. This same context dependence is preserved
when monkeys observe another individual performing these actions. Significantly,
many neurons begin to fire before the other individual actually performs a specific
action—that is, before grasping to eat as opposed to grasping to place. Thus, it seems
possible that these neurons encode not only the specific motor act but also the actor’s
intentions (Fogassi et al. 2005; see also Nakahara and Miyashita 2005; Rizzolatti and
Craighero 2004; Rizzolatti and Buccino 2005). These results are perhaps not surprising,
given the benefits of being able to predict what others are going to do.
Furthermore, like the behavioral evidence for cross-modal visual-auditory recogni-
tion, the behavioral evidence for the simultaneous recognition of identity and motives
receives support from the neurophysiological literature. In humans, one region of the
fusiform gyrus, the fusiform face area (FFA), is more engaged by human faces than by
any other visual stimulus (Kanwisher, McDermott, and Chun 1997; Kanwisher 2000).
While the FFA was originally thought to be activated only by faces, however, some
recent studies have shown that this activation depends on other factors, including
the attribution of motives to the stimuli involved. Healthy adults showed heightened
FFA activation when they were asked to observe three geometric figures engaged in
“movements intended to suggest a sense of personal agency” and “interactions that
were meant to be easily interpreted as social” (Schultz et al. 2003, 417). Subjects were
then asked to state when the “interaction” was friendly or not. Activation of the FFA
was therefore not restricted to faces but limited instead to those stimuli that had an
animate, social character. The authors conclude that the fusiform gyrus and FFA encode
The Evolution of Concepts about Agents 67

information not only about facial identity but also about the “semantic attributes of
people because of repeated perceptual experiences with faces that occur during social
interactions” (Schultz et al. 2003, 423). Making much the same argument as we have
here, the authors speculate that “the nature of the information stored in the FG might
… [include] anything that would be helpful in defining faces. … There would be a
measure of efficiency from this arrangement … that is, having representations and
computations of more abstract attributes of people interdigitated with front-end per-
ceptual processes about physical attributes” (Schultz et al. 2003, 423–424). Farah and
Heberlein (2010) review evidence that the human brain contains a dedicated system for
representing “the appearance, actions, and thoughts of people” that is innate, autono-
mous, and functions with a great deal of automaticity. This is just what we would
expect to find if natural selection had acted strongly throughout primate evolution to
favor a brain and perceptual system that innately, simultaneously, and automatically
merged information about an individual’s identity and motives.
In all the baboon experiments just mentioned, baboons that heard a vocalization
rapidly assessed the type of vocalization and the identity of the caller, and integrated
this information with their memory of recent events and their knowledge of the rela-
tionships among all the individuals involved. Based on these data, they appeared to
attribute motives to the participants and to judge the motives of different individuals
to be similar or different. When dealing with individuals perceived to share motives
(typically matrilineal kin), they responded in particular ways; when dealing with
individuals perceived not to share motives (usually unrelated animals), they reacted
differently. If the call was aggressive and came from a relative of an individual with
whom the listener had just fought, the call appeared to create in the listener’s mind the
expectation of an alliance directed against her, because the opponent and the caller,
being members of the same kin group, were assumed to have similar motivations and
were expected to act in concert. The listener therefore moved away. By contrast, if
the call was aggressive and came from an individual who was unrelated to the
previous opponent, it created no such expectation, because most alliances occur within
families. Females therefore behaved as if they assumed that the call was directed at
someone else.

3.6 Summary

For millions of years our ancestors lived in highly social groups where they saw, heard,
and interacted with the same individuals over extended periods. Under these condi-
tions, another animal’s identity becomes inseparable from its social position and its
intentions toward others. As a result, natural selection seems to have favored among
baboons and other primates a perceptual system in which the recognition of an indi-
vidual activates a rich cognitive network of information—about how the individual
68 Robert M. Seyfarth and Dorothy L. Cheney

looks and sounds, its dominance rank, its family membership, and its motivation to
behave toward specific others in particular ways. Individual recognition thus consti-
tutes a form of object perception, in which many different stimuli are bound together
to create a single coherent object—the individual—who behaves in predictable ways
toward other objects or individuals. In baboons, the predictable ways are defined in
part by the rules of close bonds among matrilineal kin and a linear dominance hierar-
chy. The object perception of others has been shaped by natural selection to be auto-
matic, effortless, and accurate—if it were not, the individual could not succeed in its
social environment.
To achieve these goals, natural selection has favored in baboons a mind that orga-
nizes data on individuals according to concepts—in the baboons’ case, concepts that
we humans label with names like matrilineal family and linear dominance rank order. The
baboons have no such names, but they do have the concepts. Not explicit concepts,
of course—we have no evidence that baboons can attach labels to different social rela-
tions, or that they are consciously aware of kin categories or their own knowledge of
them—but implicit knowledge about which animals go together, how they are likely
to interact, and the extent to which their motives are shared. Young children exhibit
implicit knowledge when they learn and remember facts but cannot explain how they
came to know them. When three-year-old children, for example, were shown the
contents of a drawer, they could easily recount the drawer’s contents at a later date,
but they could not explain how they acquired this knowledge—they explained that
they “just knew,” or that they had always known what was in the drawer (reviewed in
Nelson 2005). Children’s knowledge at this age, like that of the baboons’, is implicit,
not explicit, but no less effective for all that.

3.7 Implications

3.7.1 For Research on Theory of Mind

There is at present some controversy over the extent of a theory of mind in animals.
Much of the debate focuses on whether apes can attribute knowledge or false beliefs to
others (Call and Tomasello 2008). There are, however, many more rudimentary ways
in which animals might attribute a mental state to another, and some of these attribu-
tions may be considerably more widespread in the animal kingdom.
Baboons and other monkeys do not seem able to attribute knowledge or beliefs to
each other (Cheney and Seyfarth 2007). They do, however, seem to attribute motives
and intentions, as illustrated by the many experiments reviewed here. Indeed, there
is growing evidence that many animal species routinely attend to other individu-
als’ visual attention and intentions (e.g., Flombaum and Santos 2005; Bugnyar and
Heinrich 2005; Burkhart et al. 2012; MacLean and Hare 2012), and even distinguish
deliberate acts from accidental ones (e.g., Buttleman et al. 2007). Like young children
The Evolution of Concepts about Agents 69

(Repacholi and Gopnik 1997), monkeys appear to attribute likes, dislikes, and inten-
tions to others despite their failure to appreciate more complex mental states.

3.7.2 For Research on the Perception of Group Membership

Among social psychologists. there is considerable interest in the ways in which humans
classify others into groups, because such classifications may lead to stereotypes, preju-
dice, and conflict (e.g., Brewer 1999; Spelke and Kinzler 2007; Fiske 2010). What are
the origins of this tendency? It has been known for years that birds (Brooks and Falls
1975), primates (Cheney and Seyfarth 1982; Cheney 1987), and many other species
recognize individuals outside their own group and associate them with particular areas.
Conflict between groups, sometimes lethal, is well documented among hyenas (Smith
et al. 2010), lions (Mosser and Packer 2009), and chimpanzees (Mitani, Watts, and
Amsler 2010). Mahajan and coauthors (2011) add to this literature by showing that
semi-free-ranging rhesus monkeys associated a novel object with a specific social group
and appeared to attach a positive valence to members of their own group but a negative
valence to members of a neighboring group. These distinctions between the members
of one’s own versus the members of another group could, however, be based on famil-
iarity versus unfamiliarity, rates of interaction, or association with a particular range
or territory.
The experiments reviewed here indicate that classification of others into distinct
social units is ubiquitous in the everyday life of baboons, and indeed is essential for the
formation of social bonds, offspring survival, and longevity. This classification occurs
among individuals who live in the same group, range over the same area, see each
other every day, and interact with each other often. Perhaps most important for those
interested in comparison with humans, a baboon’s classification of others seems to be
based at least in part on the intentions she attributes to them—specifically, how she
expects them to behave toward her and toward each other.

3.7.3 For Neurophysiological Studies of Primate Cognition

We have argued that a variety of information is bound together in the perception of
primate vocalizations: information about the caller’s identity, appearance, social rela-
tionships, and behavioral intentions. This view is consistent with the hypothesis that
object knowledge—in both monkeys and humans—“may be represented in multiple
cortical areas that store information about different object attributes, such as form
… and motion” (Chao and Martin 2000, 478; see also Mahon and Caramazza 2009;
Coutanche and Thompson-Schill 2013). If this hypothesis proves correct, and if similar
mechanisms underlie the recognition of objects and individuals, then it should be pos-
sible to identify areas in the brain that are specialized not only for the identification of
familiar voices, faces, and their integration but also for the identification of appropriate
social interactions or, conversely, the detection of interactions that violate the social
70 Robert M. Seyfarth and Dorothy L. Cheney

order. The recent paper by Kumaran, Melo, and Emrah (2012) represents a first step in
this direction.


Research on baboons was supported by the National Science Foundation, the National
Institutes of Health, the Leakey Foundation, the National Geographic Society, and the
University of Pennsylvania. We thank Yale Cohen and Marc Coutanche for comments.


Adachi, I., and K. Fujita. 2007. Cross-modal representation of human caretakers in squirrel mon-
keys. Behavioural Processes 74:27–32.

Adachi, I., and R. Hampton. 2011. Rhesus monkeys see who they hear: Spontaneous cross-modal
memory for familiar conspecifics. PLoS ONE 6:e23345.

Adachi, I., H. Kuwahata, and K. Fujita. 2007. Dogs recall their owner’s face upon hearing the
owner’s voice. Animal Cognition 10:17–21.

Adolphs, R., J. A. Russell, and D. Tranel. 1999. A role for the human amygdala in recognizing
emotional arousal from unpleasant stimuli. Psychological Science 10:167–171.

Belin, P., and R. J. Zattore. 2003. Adaptation to speaker’s voice in right anterior temporal lobe.
Neuroreport 14:2105–2109.

Bergman, T., J. C. Beehner, D. L. Cheney, and R. M. Seyfarth. 2003. Hierarchical classification by

rank and kinship in baboons. Science 302:1234–1236.

Blank, H., A. Anwander, and K. von Kreigstein. 2011. Direct structural connections between
voice- and face-recognition areas. Journal of Neuroscience 31:12906–12915.

Bregman, A. S. 1990. Auditory Scene Analysis. Cambridge, MA: MIT Press.

Brewer, M. B. 1999. The psychology of prejudice: Ingroup love or outgroup hate? Journal of Social
Issues 55:429–444.

Brooks, R. J., and J. B. Falls. 1975. Individual recognition by song in white-throated sparrows. III.
Song features used in individual recognition. Canadian Journal of Zoology 53:1749–1761.

Bugnyar, T., and B. Heinrich. 2005. Ravens, Corvus corax, differentiate between knowledgeable
and ignorant competitors. Proceedings: Biological Sciences 272:1641–1646.

Burkhart, J., A. Kupferberg, S. Glasauer, and C. van Schaik. 2012. Even simple forms of social
learning rely on intention attribution in marmoset monkeys (Callithrix jacchus). Journal of
Comparative Psychology 126:129–138.

Buttleman, D., M. Carpenter, J. Call, and M. Tomasello. 2007. Enculturated chimpanzees imitate
rationally. Developmental Science 10:F31–F38.
The Evolution of Concepts about Agents 71

Call, J., and M. Tomasello. 2008. Does the chimpanzee have a theory of mind? 30 years later.
Trends in Cognitive Sciences 12:187–192.

Campanella, S., and P. Belin. 2007. Integrating face and voice in person perception. Trends in
Cognitive Sciences 11:535–543.

Cappe, C., and P. Barone. 2005. Heteromodal connections supporting multisensory integration
at low levels of cortical processing in the monkey. European Journal of Neuroscience 22:

Carey, S. 2010. The Origin of Concepts. New York: Oxford University Press.

Carey, S. 2011. The origin of concepts: A précis. Behavioral and Brain Sciences 34:116–137.

Carey, S., and E. Spelke. 1996. Science and core knowledge. Philosophy of Science 63:515–533.

Chao, L., and A. Martin. 2000. Representation of manipulable man-made objects in the dorsal
stream. NeuroImage 12:478–484.

Cheney, D. L. 1987. Interactions and relationships between groups. In Primate Societies, ed. B. B.
Smuts, D. L. Cheney, R. M. Seyfarth, R. W. Wrangham, and T. T. Struhsaker, 267–281. Chicago:
University of Chicago Press.

Cheney, D. L., and R. M. Seyfarth. 1980. Vocal recognition in free-ranging vervet monkeys.
Animal Behaviour 28:362–367.

Cheney, D. L., and R. M. Seyfarth. 1982. Recognition of individuals within and between groups
of free-ranging vervet monkeys. American Zoologist 22:519–529.

Cheney, D. L., and R. M. Seyfarth. 1990. How Monkeys See the World. Chicago: University of
Chicago Press.

Cheney, D. L., and R. M. Seyfarth. 1997. Reconciliatory grunts by dominant female baboons
influence victims’ behaviour. Animal Behaviour 54:409–418.

Cheney, D. L., and R. M. Seyfarth. 1999. Recognition of other individuals’ social relationships by
female baboons. Animal Behaviour 58:67–75.

Cheney, D. L., and R. M. Seyfarth. 2007. Baboon Metaphysics. Chicago: University of Chicago

Cheney, D. L., R. M. Seyfarth, and J. B. Silk. 1995. The responses of female baboons to anomalous
social interactions: Evidence for causal reasoning? Journal of Comparative Psychology 109:

Coutanche, M., and S. Thompson-Schill. 2013. Informational connectivity: Identifying the

co-emergence of multi-voxel patterns across the brain. Frontiers in Neuroscience 7 (15): 1–14.

Dasser, V. 1988. A social concept in Java monkeys. Animal Behaviour 36:225–230.

Emery, N. J., and D. I. Perrett. 2000. How can studies of the monkey brain help us understand
“theory of mind” and autism in humans? In Understanding Other Minds: Perspectives from Develop-
72 Robert M. Seyfarth and Dorothy L. Cheney

mental Cognitive Neuroscience, 2nd ed., ed. S. Baron-Cohen, H. Tager-Flusberg, and D. Cohen, 279–
301. Oxford: Oxford University Press.

Engh, A. L., R. R. Hoffmeier, D. L. Cheney, and R. M. Seyfarth. 2006. Who, me? Can baboons
infer the target of a vocalization? Animal Behaviour 71:381–387.

Farah, M., and A. Heberlein. 2010. Personhood: An illusion rooted in brain function? In Neuroeth-
ics, ed. M. Farah, 321–338. Cambridge, MA: MIT Press.

Fine, C., J. Lumsden, and J. R. Blair. 2001. Dissociation between “theory of mind” and executive
functions in a patient with early left amygdala damage. Brain 124:287–298.

Fiske, S. T. 2010. Social Beings: Core Motives in Social Psychology. 2nd ed. New York: Wiley.

Flombaum, J. I., and L. R. Santos. 2005. Rhesus monkeys attribute perceptions to others. Current
Biology 15:447–452.

Fogassi, L., P. F. Ferrari, B. Gesierich, S. Rozzi, F. Chersi, and G. Rizzolatti. 2005. Parietal lobe:
From action organization to intention understanding. Science 308:662–667.

Frith, C. D., and U. Frith. 2012. Mechanisms of social cognition. Annual Review of Psychology

Gallagher, H. L., and C. D. Frith. 2003. Functional imaging of “theory of mind.” Trends in Cogni-
tive Sciences 7:77–83.

Ghazanfar, A., and N. Logothetis. 2003. Facial expressions linked to monkey calls. Nature

Ghazanfar, A., J. X. Maier, K. L. Hoffman, and N. Logothetis. 2005. Multisensory integration

of dynamic faces and voices in rhesus monkey auditory cortex. Journal of Neuroscience 25:

James, W. 1890. Principles of Psychology. London: MacMillan.

Jellema, T., C. I. Baker, B. Wicker, and D. I. Perrett. 2000. Neural representation for the perception
of the intentionality of actions. Brain and Cognition 44:280–302.

Johnston, R. E., and T. A. Bullock. 2001. Individual recognition by use of odors in golden ham-
sters: The nature of individual representation. Animal Behaviour 61:545–557.

Johnston, R. E., and A. Peng. 2008. Memory for individuals: Hamsters (Mesocricetus auratus)
require contact to develop multi-component representations (concepts) of others. Journal of Com-
parative Psychology 122:121–131.

Kanwisher, N. 2000. Domain specificity in face perception. Nature Neuroscience 3:759–763.

Kanwisher, N., J. McDermott, and M. M. Chun. 1997. The fusiform face area: A module in human
extrastriate cortex specialized for face perception. Journal of Neuroscience 17:4302–4311.

Kawashima, S., M. Sugiura, T. Kato, A. Nakamura, K. Hatano, K. Ito, H. Fukuda, S. Kojima, and
K. Nakamura. 1999. The human amygdala plays an important role in gaze monitoring: A PET
study. Brain 122:779–783.
The Evolution of Concepts about Agents 73

Kitchen, D. M., D. L. Cheney, and R. M. Seyfarth. 2005. Male chacma baboons (Papio hamadryas
ursinus) discriminate loud call contests between rivals of different relative ranks. Animal Cognition

Kondo, N., E. Izawa, and S. Watanabe. 2012. Crows cross-modally recognize group members but
not non-group members. Proceedings: Biological Sciences 279:1937–1942.

Kumaran, D., H. W. Melo, and D. Emrah. 2012. The emergence and representation of knowledge
about social and nonsocial hierarchies. Neuron 76:653–666.

Leopold, D. A., and G. Rhodes. 2010. A comparative view of face perception. Journal of Compara-
tive Psychology 124:233–251.

Lorincz, E. N., T. Jellema, J.-C. Gómez, N. Barraclough, D. Xiao, and D. I. Perrett. 2005. Do mon-
keys understand actions and minds of others? Studies of single cells and eye movements. In From
Monkey Brain to Human Brain, ed. S. Dehaene, J.-R. Duhamel, M. D. Hauser, and G. Rizzolatti,
189–210. Cambridge, MA: MIT Press.

MacLean, E. L., and B. Hare. 2012. Bonobos and chimpanzees infer the target of another’s atten-
tion. Animal Behaviour 83:345–353.

Macuda, T., and W. A. Roberts. 1995. Further evidence for hierarchical chunking in rat spatial
memory. Journal of Experimental Psychology: Animal Behavior Processes 21:20–32.

Mahajan, N., M. A. Martinez, N. L. Gutierrez, G. Diesendruk, M. R. Banaji, and L. Santos. 2011.

The evolution of intergroup bias: Perceptions and attitudes in rhesus macaques. Journal of Person-
ality and Social Psychology 100:387–405.

Mahon, B. Z., and A. Caramazza. 2009. Concepts and categories: A cognitive neuropsychological
perspective. Annual Review of Psychology 6:27–51.

Mandler, G. 1967. Organization and memory. In The Psychology of Learning and Motivation:
Advances in Research and Theory, ed. K. W. Spence and J. T. Spence, 327–372. San Diego:
Academic Press.

Miller, C. T., and Y. Cohen. 2010. Vocalizations as auditory objects: Behavior and neurophysiol-
ogy. In Primate Neuroethology, ed. M. Platt and A. A. Ghazanfar, 236–254. Oxford: Oxford
University Press.

Mitani, J., D. Watts, and S. Amsler. 2010. Lethal aggression leads to territorial expansion in wild
chimpanzees. Current Biology 20:R507–R508.

Mosser, A., and C. Packer. 2009. Group territoriality and the benefits of sociality in the African
lion, Pathera leo. Animal Behaviour 78:359–370.

Nakahara, K., and T. Miyashita. 2005. Understanding intentions: Through the looking glass.
Science 308:6444–6445.

Nelson, K. 2005. Emerging levels of consciousness in early human development. In The Missing
Link in Cognition: Origins of Self-Reflecting Consciousness, ed. H. S. Terrace and J. Metcalf, 116–141.
New York: Oxford University Press.
74 Robert M. Seyfarth and Dorothy L. Cheney

Pelphrey, K. A., R. J. Viola, and G. McCarthy. 2004. When strangers pass: Processing of mutual
and averted social gaze in the superior temporal sulcus. Psychological Science 15:598–603.

Penn, D., K. Holyoak, and D. Povinelli. 2008. Darwin’s mistake: Explaining the discontinuity
between human and nonhuman minds. Behavioral and Brain Sciences 31:109–178.

Perry, S., H. C. Barrett, and J. Manson. 2004. White-faced capuchin monkeys show triadic aware-
ness in their choice of allies. Animal Behaviour 67:165–170.

Petkov, C. I., C. Kayser, T. Steudel, K. Whittingstall, M. Augath, and N. K. Logothetis. 2008. A

voice region in the monkey brain. Nature Neuroscience 11:367–374.

Proops, L., K. McComb, and D. Reby. 2009. Cross-modal individual recognition in domestic
horses (Equus caballus). Proceedings of the National Academy of Sciences of the United States of America

Rendall, D., P. S. Rodman, and R. E. Emond. 1996. Vocal recognition of individuals and kin in
free-ranging rhesus monkeys. Animal Behaviour 51:1007–1015.

Repacholi, B., and A. Gopnik. 1997. Early reasoning about desires: Evidence from 14- and
18-month olds. Developmental Psychology 33:12–21.

Rizzolatti, G., and G. Buccino. 2005. The mirror neuron system and its role in imitation and
language. In From Monkey Brain to Human Brain, ed. S. Dehaene, J.-R. Duhamel, M. D. Hauser, and
G. Rizzolatti, 189–210. Cambridge, MA: MIT Press.

Rizzolatti, G., and L. Craighero. 2004. The mirror-neuron system. Annual Review of Neuroscience

Santos, L. R., J. I. Flombaum, and W. Phillips. 2006. The evolution of human mindreading: How
non-human primates can inform social cognitive neuroscience. In Evolutionary Cognitive
Neuroscience, ed. S. M. Platek, J. P. Keenan, and T. K. Shackelford, 433–456. Cambridge, MA: MIT

Schino, G., B. Tiddi, and E. Polizzi di Sorrentino. 2006. Simultaneous classification by rank and
kinship in Japanese macaques. Animal Behaviour 71:1069–1074.

Schultz, R. T., D. J. Grelotti, A. Klin, J. Kleinman, C. Van der Gaag, R. Marois, and P. Skudlarski.
2003. The role of the fusiform face area in social cognition: Implications for the pathobiology of
autism. Philosophical Transactions of the Royal Society of London: Series B, Biological Sciences

Schusterman, R. J., and D. A. Kastak. 1998. Functional equivalence in a California sea lion: Rele-
vance to animal social and communicative interactions. Animal Behaviour 55:1087–1095.

Seyfarth, R. M., and D. L. Cheney. 2001. Cognitive strategies and the representation of social rela-
tionships by monkeys. In Evolutionary Psychology and Motivation, Nebraska Symposium on Motiva-
tion, vol. 48, ed. J. A. French, A. C. Kamil, and D. W. Leger, 145–178. Lincoln: University of
Nebraska Press.
The Evolution of Concepts about Agents 75

Silk, J. B. 1999. Male bonnet macaques use information about third-party rank relationships to
recruit allies. Animal Behaviour 58:45–51.

Silk, J. B., S. Alberts, and J. Altmann. 2003. Social bonds of female baboons enhance infant
survival. Science 302:1331–1334.

Silk, J. B., J. C. Beehner, T. Bergman, C. Crockford, A. Engh, L. Moscovice, R. Wittig, R. M.

Seyfarth, and D. L. Cheney. 2010a. Female chacma baboons form strong, equitable, and enduring
social bonds. Behavioral Ecology and Sociobiology 64:1733–1747.

Silk, J. B., J. C. Beehner, T. Bergman, C. Crockford, A. Engh, L. Moscovice, R. Wittig, R. M.

Seyfarth, and D. L. Cheney. 2010b. Strong and consistent social bonds enhance the longevity of
female baboons. Current Biology 20:1359–1361.

Silk, J. B., J. C. Beehner, T. Bergman, C. Crockford, R. M. Wittig, A. L. Engh, R. M. Seyfarth, and

D. L. Cheney. 2009. The benefits of social capital: Close bonds among female baboons enhance
offspring survival. Proceedings: Biological Sciences 276:3099–3104.

Silk, J. B., D. L. Cheney, and R. M. Seyfarth. 1996. The form and function of post-conflict interac-
tions between female baboons. Animal Behaviour 52:259–268.

Sliwa, J., J. R. Duhamel, O. Pascalis, and S. Wirth. 2011. Spontaneous voice-face identity match-
ing by rhesus monkeys for familiar conspecifics and humans. Proceedings of the National
Academy of Sciences of the United States of America 108:1735–1740.

Smith, J. E., R. VanHorn, K. S. Powning, A. R. Cole, K. E. Graham, S. K. Memenis, and

K. Holekamp. 2010. Evolutionary forces favoring intragroup coalitions among spotted hyenas
and other animals. Behavioral Ecology 21:284–303.

Spelke, E., and K. Kinzler. 2007. Core knowledge. Developmental Science 10:89–96.

Tibbetts, E. A., and J. Dale. 2007. Individual recognition: It is good to be different. Trends in
Ecology & Evolution 22:529–537.

Tsao, G. Y., W. A. Friewald, R. B. Tootell, and M. S. Livingston. 2006. A cortical region consisting
entirely of face-selective cells. Science 311:670–674.

von Kriegstein, K., A. Kleinschmidt, P. Sterzer, and A. L. Giraud. 2005. Interaction of face and
voice areas during speaker recognition. Journal of Cognitive Neuroscience 17:367–376.

Wittig, R. M., C. Crockford, R. M. Seyfarth, and D. L. Cheney. 2007. Vocal alliances in chacma
baboons, Papio hamadryas ursinus. Behavioral Ecology and Sociobiology 61:899–909.

Wittig, R. M., C. Crockford, E. Wikberg, R. M. Seyfarth, and D. L. Cheney. 2007. Kin-mediated

reconciliation substitutes for direct reconciliation in baboons. Proceedings: Biological Sciences
II Concepts and the Brain
4 Missed Connections: A Connectivity-Constrained Account of the
Representation and Organization of Object Concepts

Bradford Z. Mahon

4.1 Introduction

One of the most exciting open issues in the cognitive and brain sciences is how
conceptual knowledge is represented and organized in the brain. A recent prolifera-
tion of methods that allow cognition to be studied in vivo in healthy subjects has
accelerated our understanding of how the brain processes conceptual knowledge. Two
broad goals can be framed for the empirical study of concepts: a scientific goal and a
clinical goal.
The scientific goal is to develop a model of how concepts are organized and rep-
resented, at both the cognitive and neural levels. Such a model would specify the
format of conceptual information and the processing dynamics that govern how that
information is retrieved according to current task demands. For instance, the types of
information that are prioritized are different if one looks at a cup with the intention
of taking a drink from it, versus with the intention of identifying it as being “my cup,”
versus checking whether it is empty. A model would specify how the same visual input
in all such situations is “routed” through sensorimotor and conceptual systems. This
would involve specifying what information is necessary for the task, versus relevant
but not necessary, versus activated but entirely ancillary, as well as the order in which
information is retrieved. Such a model would specify how conceptual knowledge inter-
faces with other cognitive systems, such as planning and executive function, linguistic
processing, as well as how it interfaces with sensorimotor input and output systems.
Finally, such a model would also specify not only the structure and processing dynam-
ics of the conceptual system at a cognitive level, but how that system is organized and
distributed in the brain.
The clinically oriented goal can be separated into prognostic and treatment-oriented
components. On the basis of a working model of how concepts are organized and
represented, a prognostic clinical goal is, for instance, to develop accurate means for
determining expected outcomes after acute brain injury, or the trajectory of loss of
80 Bradford Z. Mahon

conceptual information in the context of progressive degenerative diseases such as

Alzheimer’s disease or semantic dementia. Another clinically oriented goal is to use
a working model of conceptual organization to help guide neurosurgical planning
of the location and extent of cortical resections. Finally, we would want to use our
understanding of how concepts are organized and represented in the brain to guide
rehabilitation of lost function, where rehabilitation can include existing cognitive-
behavioral therapy and prospective approaches that may seek to actually repair or
replace damaged tissue.
To date, the study of concepts in the brain has generally focused on analyses
of the factors that modulate processing local to a particular region of the conceptual
system—for instance, the visual factors that modulate visual object recognition, or the
motor factors that modulate object-directed action. This local, bottom-up, approach has
been inherited from well-established traditions in neurophysiology and psychophys-
ics, where it has been enormously productive for mapping psychophysical continua
in primary sensory systems. Here I argue, however, that the same approach will not
yield equally useful insights for understanding the principles that determine the orga-
nization and representation of conceptual knowledge. The reason is that unlike the
peripheral sensory systems, the patterns of neural responses that reflect conceptual
processing are only partially driven by the physical input—they are also driven by
how the stimulus is interpreted, and that interpretation does not occur in a single,
isolated region. Thus, a critical step for developing a model of conceptual organization
and representation is to articulate how multiple sources of information are integrated
in real time and how concepts interface with other cognitive and sensorimotor
systems. This means that to move forward, as a field, the connectivity of the concep-
tual system with language, executive, sensory, motor, and other systems must become
the new “unit of analysis.” Connectivity is not just wiring; the connections are not
passive conduits through which information is passed. The connectivity of the system
constrains the order in which information is accessed and can be weighted, and that
weighting of information is a central aspect of the computations that form conceptual
The current state of the field, summarized below, is characterized by a somewhat
more static notion of concepts and has developed largely by setting issues of connec-
tivity aside. As will be seen, a tremendous amount of progress has been made in recent
decades. But the depth of our understanding of both old and new theoretical questions
will quickly reach asymptote without a shift in emphasis toward understanding the
role played by connectivity. In keeping with the theme of this volume, this chapter is
my attempt to shoot an azimuth of where we are headed as a field; I do this by outlin-
ing a theoretical framework that places connectivity at the center of what needs to be
The Representation and Organization of Object Concepts 81

4.2 Overview of the Hypothesis Space

A distinction can be drawn between hypotheses about concept organization and

hypotheses about concept representation. Hypotheses about how concepts are orga-
nized in the brain are typically concerned with the causes of an observed physical
distribution of conceptual information in different regions of cortex and the reasons
the conceptual system is observed to fractionate along the lines that it does under
conditions of brain damage. Hypotheses about how concepts are represented are
concerned with the representational format of conceptual information. For instance, a
representational issue is whether conceptual knowledge is represented in a modality-
specific format (visual, motor) or in an abstract format. As might be expected, the
distinction between representation and organization is blurred by theories that make
claims about, or have implications for, both the organization and the representation
of concepts. Nevertheless, it is useful to draw a distinction between organization and
representation, principally because there exists an asymmetry in the types of confirma-
tory empirical evidence that have been marshaled in support of theories that focus on
representation versus theories that focus on organization.
The literature review below is fast and loose, and heavily curated; it is entirely in
the service of motivating a shift in perspective toward studying the connectivity of
the conceptual system. It also makes little attempt to be ecumenical and is guided by
a theoretical framework that has been developed over the past ten years with Alfonso
Caramazza (Caramazza and Mahon 2003, 2006; Mahon and Caramazza 2003, 2005,
2008, 2009, 2011). An array of excellent reviews from other theoretical perspectives
that discuss a broader range of findings can be found in Barsalou (1999); Binder and
Desai (2011); Borgo and Shallice (2003); Chatterjee (2010); Cree and MacRae (2003);
Gallese and Lakoff (2005); Glenberg, Sato, and Cattaneo (2008); Grill-Spector and
Malach (2004); Hart et al. (2007); Humphreys and Forde (2001); Johnson-Frey (2004);
Kemmerer et al. (2012); Kemmerer (forthcoming); Kiefer and Pulvermüller (2012); Laws
(2005); Lewis (2006); Martin (2007); Op de Beeck, Haushofer, and Kanwisher (2008);
Patterson, Nestor, and Rogers (2007); Pülvermuller (2005); Sartori and Lombardi (2004);
Simmons and Barsalou (2003); Thompson-Schill (2003); Tyler and Moss (2001); Vinson
et al. (2003).
The scope of the argument to follow is restricted to the representation and orga-
nization of object concepts. This leaves most of conceptual space unaccounted for,
including concepts expressed as nouns that do not have physical referents (e.g., DREAM,
GOAL, PIETY, etc.), as well as concepts that apply to actions, abstract verbs, and many

other types of concepts, such as numbers, theory of mind, moral concepts, and logi-
cal concepts and reasoning (among other domains as well). The restricted scope of
the review helps to gain a solid, if limited, footing on the theoretical issues pertaining
82 Bradford Z. Mahon

to the organization and representation of object concepts. A complete theory would

presumably have within its scope all domains of conceptual processing; whether it will
be possible to develop such an account is an empirical and methodological challenge
that we face as a field.

4.3 Associative Evidence and Theories of the Representation of Concepts

A widely discussed theoretical framework about the representation of conceptual

knowledge is the “embodied cognition hypothesis.” The central idea of this framework,
applied to concepts, is that sensorimotor representations are reactivated or “simulated”
in the course of conceptual analysis, and that sensorimotor activation is a necessary
and intermediary step in the computation of meaning (see Allport 1985 for an early
formulation of this view). The strong (and arguably most interesting) form of this view
is that there is no representational distinction between conceptual information and
sensorimotor information: retrieving concepts consists of simulation or reactivation of
sensorimotor information that is/was activated, either when we initially acquired the
concept or when we interact with instantiations of that concept. Because this is the
central claim of the hypothesis, the theory is committed to the view that conceptual
retrieval involves the retrieval of stored sensorimotor information. These sensorimotor
“memories” are sensory or motor in their format, and therefore they are assumed to
be “token-based” representations, or representations of actual instances of sensory or
motor experiences. Another way to think about such theories is that they are a type of
exemplar-based model of semantic memory, where the exemplars consist of sensorimo-
tor information and are in a sensory or motor format.
An important type of evidence argued to support the embodied cognition hypoth-
esis consists of demonstrations that motor processes are automatically engaged when
participants perform conceptual and perceptual tasks that do not require, on a logical
analysis of the task, overt activation of the motor system. Such motor activation has
been observed in functional neuroimaging, neurophysiological recordings in nonhu-
man primates and in humans, EEG, behavior, transcranial magnetic stimulation (TMS),
and kinematic analyses (for empirical reviews and theoretical discussions, see, e.g.,
Barsalou et al. 2003; Boulenger et al. 2006; Gallese and Lakoff 2005; Martin 2007;
Pulvermüller 2005; Rizzolatti and Craighero 2004). For instance, Hauk and colleagues
(2004) found overlap in the regions of the motor cortex that were activated for both
physical actions and words that describe actions (e.g., kick). Foot-related action words
like kick activated dorsal parts of the motor cortex while hand-related action words
like pick activated more lateral and ventral parts of the motor cortex, following the
known pattern of somatotopy. As another example, Glenberg, Sato, and Cattaneo
(2008) found that when participants moved hundreds of beans from a container near
them to a container farther away, they were slower to judge sentences as sensible that
The Representation and Organization of Object Concepts 83

described action events in which objects were moved away from the body. Glenberg
and colleagues argued that fatiguing the motor system selectively interfered with the
comprehension of sentences whose meaning implied a directionality congruent with
the direction of the prior bean movements.
Recent interest in the embodiment of concepts parallels recent interest in motor
theories of perception. The original motor theory of speech perception (Liberman et al.
1967; Liberman and Mattingly 1985) stated that speech recognition was fundamentally
a process of recognizing the motor actions (tongue/articulatory movements) of the
speaker and not one of recognizing the auditory perceptual information per se. Thus,
speech recognition consisted of simulating the motor output programs that would be
necessary to produce the sounds being recognized. The motor theory of speech percep-
tion, and more recently, the motor theory of action recognition, have enjoyed a renais-
sance because of the discovery of so-called mirror neurons: neurons in premotor and
other motor-relevant structures in macaques that fire both when the monkey performs
an action and when it observes another individual (human, monkey) performing an
action (e.g., di Pellegrino et al. 1992; Gallese et al. 1996). Mirror neurons are thought to
provide the empirical substrate for a reformulated motor theory of action perception,
and thus provide independent evidence for the notion that processing in one system
(perception) is in part constituted by (i.e., involves as a necessary and intermediary
step) processing in the motor system (for critical discussion, see Binder and Desai 2011;
Chatterjee 2010; Dinstein et al. 2008; Hickok 2009, 2010; Hickok et al. 2008; Hickok
et al. 2011; Lingnau, Gesierich, and Caramazza 2009; Mahon and Caramazza 2005,
2008; Stasenko, Garcea, and Mahon, 2013).
The critical issue is whether demonstrations that the motor system is activated
during perceptual or conceptual analysis indicate, as presumed by the embodied
cognition hypothesis and the motor theory of action recognition, that motor informa-
tion plays a constitutive (i.e., necessary) role in perceptual or conceptual analysis. The
alternative is that activation spreads from perceptual or conceptual levels of process-
ing through to motor processes. There are different ways in which such an alternative
could be formulated. For instance, it could be argued, in the context of motor activa-
tion during perception, that the dynamics of the sensorimotor systems are such that
activation propagates to the motor system only after the stimulus has been recognized
as such. Alternatively, it could be argued that activation cascades forward to the motor
system from input levels of processing prior to completion of processing at those input
levels. The broader point is that a range of alternative accounts can be formulated
to explain why the motor system is activated during perception (for discussion, see
Mahon and Caramazza 2008; Stasenko, Garcea, and Mahon, 2013).
By analogy, there is evidence that the phonology of words that are never overtly
produced, but that are semantically related to actually produced words, is activated in
the course of speech production (e.g., Costa, Caramazza, and Sebastián-Gallés 2000;
84 Bradford Z. Mahon

Peterson and Savoy 1998). It has never been argued, however, that such observations
sanction the inference that the activated phonological information constitutes, even in
part, the semantics of the unproduced words. Rather, the debate concerns the dynamics
of information flow within the speech production system, and whether it is cascaded
activation or serial and discrete. The relationship between evidence and theory exactly
mirrors the relationship between observations of motor activation during perceptual or
conceptual processing, and the embodied cognition hypothesis. Thus the implication
is that the representational inferences that have been argued to support the embodied
cognition hypothesis are, at best, premature, and the available evidence is, at best,
(only) consistent with the embodied cognition hypothesis.

4.4 Patient Evidence and the Embodied Cognition Hypothesis

Recent work with brain-damaged patients highlighted cases where conceptual and
motor abilities are seen to be impaired together—that is, theoretically interesting asso-
ciations of impairments. For instance, Pazzaglia, Pizzamiglio, and colleagues (2008)
observed that patients with buccofacial apraxia (impairments producing sounds related
to facial and mouth structures, e.g., whistling or slurping a straw) had greater difficulty
recognizing mouth action sounds compared with hand action sounds, whereas patients
with limb apraxia (difficulties performing skilled actions like pounding a hammer) had
greater difficulty recognizing limb action sounds compared with mouth action sounds
(see Mahon 2008 for discussion; see also Pazzaglia, Pizzamiglio, et al. 2008). Buxbaum
and colleagues (2005) found an association at the group level in the ability of patients
to imitate certain types of actions and their ability to recognize actions (see Negri
et al. 2007 for replication, extension, and critical discussion). Boulenger and colleagues
(2008) combined a masked priming paradigm with a lexical decision task to study
semantic priming effects in a nondemented group of Parkinson’s patients (n = 10) who
were either off or on dopaminergic treatment. It is known that Parkinson’s patients
show relative inactivation of motor cortices when they are off, compared with when
they are on, dopaminergic treatment. The authors found that the magnitude of the
masked priming was modulated according to whether the patients were on or off their
medication, and importantly, this modulation was present only for action word targets
and not for concrete nouns. Another recent and rich example is the study by Bonner
and Grossman (2012), who found that patients with logopenic variant of primary pro-
gressive aphasia, a variant that leads to cortical atrophy first around Hershel’s gyrus,
were impaired for knowledge about the typical sounds that objects make.
However, there is also dissociative patient evidence indicating that action produc-
tion can be impaired while action recognition is spared, both in the domain of hand
actions (Negri et al. 2007; Rumiati et al. 2001; Rapcsak et al. 1995; for reviews, see
Mahon and Caramazza 2005, 2008) and in the domain of speech perception (Hickok
et al. 2011; Rogalsky et al. 2011; for reviews, see Hickok 2009, 2010; Stasenko, Garcea,
The Representation and Organization of Object Concepts 85

and Mahon, 2013; Toni et al. 2008). For instance, patients with apraxia of object use
are not necessarily impaired for naming the same objects or retrieving function knowl-
edge about those objects (e.g., Buxbaum, Veramonti, and Schwartz 2000; Buxbaum
and Saffran 2002; Ochipa, Rothi, and Heilman 1989; Rapcsak et al. 1995; for review, see
Mahon and Caramazza 2005).
The existence of dissociative evidence is problematic for the claim that motor
information forms a necessary component of conceptual or perceptual processing. The
dissociative patient evidence is sufficient to reject strong forms of the embodied con-
cept hypothesis and strong forms of the motor theory of action perception (for dis-
cussion, see Garcea and Mahon 2012; Mahon and Caramazza 2005, 2008; Stasenko,
Garcea, and Mahon, 2013; Hickok 2010; Chatterjee 2010). This conclusion carries
with it the burden of explaining (1) why the motor system is activated during
(perceptual and conceptual) tasks that do not, on a logical analysis of what is involved
in those tasks, necessitate overt activation of motor information, and (2) why the
above-described associative patient evidence is observed (e.g., Buxbaum, Kyle, and
Menon 2005; Bonner and Grossman 2012; Pazzaglia, Pizzamiglio, et al. 2008, Pazzaglia,
Smania, et al 2008). One possibility is that activation spreads from perceptual or con-
ceptual levels of representation to the motor system. In the context of the embodied
cognition hypothesis, the question arises as to what function such activation might
serve. The associative patient evidence suggests that such activation is not irrelevant
or ancillary but that it may play some, as yet unspecified, function. I return to these
issues below.
The broader point is that the core issue that must be elucidated concerns how
information is exchanged among sensory, motor, and conceptual systems. What we
are missing, as a field, is a theory of the dynamics of activation spread between percep-
tual or conceptual processes and the motor system. Only in the context of a specific
theory of the dynamics of sensorimotor and conceptual processing can strong infer-
ences about the representational format of concepts be derived from observations that
the motor system is automatically activated during perceptual or conceptual tasks.

4.5 Dissociative Evidence and Theories of the Organization of Concepts

One of the most intriguing of neuropsychological phenomena are category-specific

semantic deficits. Category-specific semantic deficits are impairments to conceptual
knowledge that differentially, or selectively, affect information from one semantic
category. Figure 4.1A (plate 5) shows the picture-naming performance from some
well-studied patients and represents the full range of semantic categories that can be
differentially or selectively impaired in patients with category-specific semantic defi-
cits. The categories of category-specific semantic deficits are living animate (animals),
living inanimate (fruits/vegetables), conspecifics, and tools (for an exhaustive review of
the empirical literature through 2001, see Capitani et al. 2003).
86 Bradford Z. Mahon

Figure 4.1 (plate 5)

Patients with category-specific semantic deficits may be differentially, or even selectively,
impaired for knowledge of animals, plants, conspecifics, or artifacts. The knowledge impairment
cannot be explained in terms of a differential impairment to a sensory or motor-based modal-
ity of information. While discussion and debate continues as to whether noncategorical dimen-
sions of organization may lead to category-specific brain organization, there is consensus that
the phenomenon itself is categorical. (A) Picture-naming performance of patients studied with
materials that were carefully balanced to equate various continuous dimensions across catego-
ries (e.g., frequency, familiarity, visual complexity). The four major patterns of category-specific
semantic deficits are represented. (B) Semantic attribute question performance for six represen-
tative patients with differential impairments for living animate. As shown across the patients,
impairments for a category are associated with impairments for all types of knowledge about items
from that category. Figure reproduced from Mahon and Caramazza (2011), with permission.
The Representation and Organization of Object Concepts 87

Category-specific semantic deficits have been particularly fertile ground for the devel-
opment and evaluation of hypotheses about how conceptual information is organized.
The general question that these theories seek to answer is, How is knowledge orga-
nized in the normal system such that damage can lead to impairments that respect
semantic category distinctions? Current theories can be separated into reductionist
and nonreductionist approaches; reductionist theories do not posit semantic category
as an organizing principle in the mind/brain, while nonreductionist theories do
posit semantic category as an organizing principle. Within reductionist theories, a
further distinction can be made between eliminativist reductionist and non-eliminativist
reductionist, with eliminativist approaches denying any principles of neural organization
whatsoever, and non-eliminativist making clear positive proposals about neural organi-
zation (but not appealing to semantic category or domain as an organizing principle).

4.5.1 Eliminativist Reductionist Theories

It was recognized early on in the study of category-specific semantic deficits that the
impairments in some patients could be explained because items from different seman-
tic categories tended to differ along certain variables, such as lexical frequency, con-
cept familiarity, or the visual complexity of the images that were used to establish
the presence of impairments. Thus, if items were sampled without attention to those
variables, then some categories within the neuropsychological tests might be “easier”
than other categories, thus leading to spurious category dissociations (see Funnell and
Sheridan 1992; for recent discussion and for extensive normative work, see Barbarotto
et al. 2002; Cree and MacRae 2003). More recently, Sartori, Lombardi, and colleagues
(Mechelli et al. 2006; Sartori and Lombardi 2004; Sartori, Lombardi, and Mattiuzzi
2005) developed a measure termed semantic relevance, which is a nonlinear combina-
tion of the frequency with which particular features are produced for an item and the
distinctiveness of that feature across all concepts in the database. In addition, the issue
of whether items from some categories are more visually complex than others can
be expressed as a cognitive issue and not merely an issue about the stimuli over
which patients are tested. It may be argued that the structure of the world, and our
representation of that structure, is such that the visual knowledge of one category
is more “tightly packed” or “crowded” than other categories. Arguments differ as to
which categories have representations that are more densely packed, but proposals
agree that a higher density of representations renders them more susceptible to impair-
ment (e.g., Humphreys and Forde 2001; Laws 2005).1

1. It should be noted, however, that it is not obvious that higher density (i.e., higher within-
category similarity) should lead to greater susceptibility to impairment, as, for instance, Tyler,
Moss, and colleagues (Tyler et al. 2000; Tyler and Moss 2001) argued in a somewhat different
context that high correlations among shared features confer resistance to damage (but see Devlin
et al. 2002).
88 Bradford Z. Mahon

In summary, a number of dimensions may vary across semantic categories that are
represented at both conceptual and nonconceptual levels of processing (e.g., lexical
frequency, concept familiarity, visual/structural similarity). The critical test of such
accounts, as accounts of the existence of the phenomenon of category-specific defi-
cits, is (1) whether category dissociations remain when items are carefully matched
across categories (for discussion, see Capitani et al. 2003), and (2) whether double dis-
sociations can be observed over the same materials across patients (e.g., Caramazza
and Hillis 1991). Because both (1) and (2) have been answered in the affirmative, we
can conclude that category-specific semantic deficits are not spurious, that is, cannot
be reduced to such uncontrolled dimensions. This conclusion rules out the broad class
of eliminativist reductionist theories (for evidence and arguments, see Caramazza and
Mahon 2003; Mahon and Caramazza 2009).

4.5.2 Non-eliminativist Reductionist Theories

The modern study of concepts in the brain was inaugurated by the empirical and theo-
retical work of Warrington, McCarthy, and Shallice (Warrington and McCarthy 1983,
1987; Warrington and Shallice 1984). In a series of papers, they documented the first
well-described cases of category-specific semantic deficits and proposed the most influ-
ential theory that is still widely discussed today: the sensory/functional theory.
The sensory/functional theory makes two assumptions:

1. First, the semantic system is organized by modality or type of information. The

original distinction drawn was between visual-perceptual and functional-associative
systems (see Lhermitte and Beauvois 1973 for earlier work on this assumption). More
recently, Crutch and Warrington (2003; see also Cree and MacRae 2003; Vigliocco et al.
2004) proposed that the semantic system may be more finely structured by modality
or type of information. For instance, visual-perceptual can be further fractionated into
color, form, and surface properties (Cree and MacRae 2003; Crutch and Warrington
2. Second, the ability to recognize items from different semantic categories differen-
tially depends on different modalities or types of information. For instance, the ability
to recognize and identify animals, it was argued, differentially depends on visual-
perceptual information, while the ability to recognize tools and other man-made arti-
facts differentially depends on functional-associative knowledge. The comprehensive
semantic feature norming work of Cree and MacRae (2003; see also Vigliocco et al.
2004) is largely directed at this second assumption. Cree and MacRae asked healthy
subjects to produce features of common objects, and the resulting features were
taxonimized into nine knowledge types that could have plausible neural bases. The
authors then used clustering methods to argue that some knowledge types were more
important or salient for some semantic categories.
The Representation and Organization of Object Concepts 89

The two core assumptions of the sensory/functional theory together explain

category-specific semantic deficits as arising from damage to a modality or type of
knowledge on which successful identification of items from the impaired category dif-
ferentially depends. For this reason, the theory is non-eliminativist (it makes a strong
and positive claim about neural organization) but is reductionist with respect to seman-
tic categories (as it posits the relevant underlying organizational principle is modality
rather than category).
It is important to note that the sensory/functional theory is not committed to a
particular view about the format of conceptual representations—that is, the sensory/
functional theory could, or could not, be formulated as an embodied cognition hypoth-
esis. The hypothesis that semantic information can be distinguished by modality or
type could be proposed as a claim about the content of semantic information and not its
format (for discussion, see Caramazza et al. 1990). This is a point that is often obscured
in the literature, where the sensory/functional theory is assumed to be a claim about
the format of conceptual knowledge and thus run together with embodied proposals.
However, one can accept a strong representational distinction between concepts and
sensorimotor processes (i.e., conceptual information is dissociable from sensorimotor
processes), and still argue that modality or type of information is the principle dimen-
sion along which concepts are organized. In fact, this was the original proposal by
Warrington and her collaborators. Subsequent developments of the theory (e.g., Martin,
Ungerleider, and Haxby 2000) have argued for more nuanced positions, for instance,
that object concepts are stored adjacent to the sensory and motor systems that were
active when the concept was acquired. Even the (so-called) sensory/motor theory of
Martin and collaborators (see also Martin 2007), however, is not committed to the
view that the format of conceptual representations is strictly sensorimotor. Still other
hypotheses argue for somewhat stronger marriages between the sensory/functional
theory and embodied views of concepts (see, for instance, Simmons and Barsalou 2003).
A number of arguments have been raised against the sensory/functional theory.
On the one hand, the evidence marshaled in support of the assumption that different
categories differentially rely on different types or modalities of knowledge for their
recognition has been questioned on methodological grounds (e.g., Caramazza and
Shelton 1998). More recent investigations (e.g., Cree and MacRae 2003) largely over-
came the methodological limitations that attended earlier studies. The findings of the
more recent and sophisticated normative studies, however, are not obviously relevant
to the key assumption that different types of knowledge are differentially important
for distinguishing within categories. For instance, Cree and MacRae showed that differ-
ent types of knowledge are important for distinguishing among different categories, or
one category from other categories. In one finding, the authors showed that color is
important for distinguishing fruits and vegetables from other categories, while biologi-
cal motion was important for distinguishing animals from other categories. Patients
90 Bradford Z. Mahon

with category-specific semantic deficits, however, have difficulties distinguishing not

between categories but within categories. Thus, it is not obvious that there is in fact
evidence for the assumption that different types or modalities of information are dif-
ferentially important for distinguishing among items within categories.
Another argument against the sensory/functional theory, and perhaps the most
damaging argument, is that patients with category-specific semantic deficits do not
present with differential impairments for the modality or type of knowledge on which
the impaired category (putatively) depends (see figure 4.1B, plate 5). In other words,
patients with disproportionate impairments for animals do not have a corresponding
disproportionate impairment for visual-perceptual knowledge. Similarly, patients with
disproportionate impairments for visual-perceptual knowledge do not necessarily have
a disproportionate impairment for animals. Such dissociations between the category
specificity of the deficit and the modality specificity of the deficit are the norm, rather
than the exception (see Capitani et al. 2003).

4.5.3 Nonreductionist Theories

According to the nonreductionist view, category-specific semantic deficits arise because
the damage affects a brain region or network of brain regions that is devoted to a par-
ticular semantic domain of knowledge. The domains for which there are specialized
systems are limited to those that could have had an evolutionarily significant history
(living animate, living inanimate, conspecifics, and tools). This hypothesis was initially
articulated in the context of category-specific semantic impairments by Caramazza
and Shelton (1998; see also, e.g., Capitani et al. 2003; Farah and Rabinowitz 2003;
Samson and Pillon 2003). Subsequent formulations of the domain-specific hypothesis
(Caramazza and Mahon 2003, 2006; Mahon and Caramazza 2009, 2011) have empha-
sized that the semantic system is also organized by modality or type of information.
In other words, there may be two orthogonal dimensions of organizations, perhaps
hierarchically structured: domain and modality. I return to this issue below.
In summary, the picture that emerges from the last several decades of work on
category-specific semantic deficits is as follows: (1) Category-specific semantic deficits
survive stringent control of stimulus variables and are observed to doubly dissociate
across patients tested with the same set of materials. These facts rule out, broadly speak-
ing, theories that posit that the phenomenon arises because a dimension (e.g., visual
complexity, relevance) is correlated with a semantic category distinction. (2) Category-
specific semantic deficits affect all types of knowledge that have been tested, indicating
that a deficit to a particular modality or type of knowledge cannot explain the exis-
tence of the phenomenon (figure 4.1B, plate 5). In this context, we (e.g., Caramazza
and Mahon 2003; Mahon and Caramazza 2009) have concluded that semantic domain
is an organizing principle of conceptual knowledge of objects, and that the most attrac-
tive model (for other reasons, see below) has at least two orthogonal dimensions of
organization: semantic domain and modality or type of information.
The Representation and Organization of Object Concepts 91

4.6 Functional MRI Evidence for the Constraints That Shape Object Knowledge
in the Brain

4.6.1 Ventral and Dorsal Object-Processing Streams

An important development in cognitive neuroscience that has paralleled the articu-
lation of theories of semantic organization is the discovery of multiple channels of
visual processing (Goodale and Milner 1992; Ungerleider and Mishkin 1982). It is
now known that cortical visual processing bifurcates into two independent but inter-
connected streams (for discussion of how best to characterize the two streams, see
Pisella et al. 2006; Schenk 2006; see also Merigan and Maunsell 1993). The ventral
object-processing stream projects from V1 through the ventral occipital and temporal
cortices, terminating in anterior regions of the temporal lobe, and subserves visual
object identification. The dorsal object-processing stream projects from V1 through
the dorsal occipital cortex to the posterior parietal cortex, and subserves object-directed
action and spatial analysis for the purpose of object-directed grasping. The two visual
systems hypothesis was initially formulated on the basis of neuropsychological evi-
dence, in which ventral lesions led to impairments for perception and identification
of object attributes but spared action toward the same objects, while dorsal lesions led
to action impairments that spared perception (e.g., Goodale et al. 1991; Pisella et al.
2000; Ungerleider and Mishkin 1982). There has since been an enormous amount of
imaging work confirming the distinction between ventral and dorsal object-processing
streams (e.g., Binkofski et al. 1998; Culham et al. 2003; Mahon et al. 2007; Shmuelof
and Zohary 2005).

4.6.2 Category Specificity in the Ventral Object-Processing Stream

There is a vibrant literature studying category specificity in humans using functional
magnetic resonance imaging (fMRI). The most widely studied categories in high level
visual regions, and the categories for which specific regions of the brain exhibit dif-
ferential blood oxygen level dependent (BOLD) responses are faces, animals, body
parts, tools, places, and words (for reviews, see Bookheimer 2002; Gerlach 2007;
Grill-Spector and Malach 2004; Martin 2007; Op de Beeck, Haushofer, and Kanwisher
2008; Thompson-Schill 2003). On the ventral surface of the temporo-occipital cortex,
in the ventral object-processing stream, there is a consistent topography by seman-
tic category across individuals. For instance, viewing tools leads to differential BOLD
contrast in the left medial fusiform gyrus, while viewing animate living things (ani-
mals and faces) leads to differential BOLD contrast in the lateral fusiform gyrus (Chao,
Haxby, and Martin 1999; Kanwisher, McDermott, and Chun 1997; for earlier work,
see Allison et al. 1994). The region of the lateral fusiform gyrus that exhibits larger
responses to faces compared with a range of other categories (Downing et al. 2006)
has been named the fusiform face area (FFA). The face area tends to be lateralized (or
biased) toward the right hemisphere, while often in the homologous region of the left
92 Bradford Z. Mahon

hemisphere, selectivity for printed words is observed (for review, see Dehaene et al.
2005; for modeling work and discussion of this asymmetry, see Plaut and Behrmann
2011). Place stimuli, such as houses or scenes, differentially drive BOLD responses in a
more anterior and medial location in the ventral stream adjacent to the hippocampus,
called the parahippocampal gyrus (the region called the parahippocampal place area,
or PPA; Epstein and Kanwisher 1998; see also Bar and Aminoff 2003). Finally, there
are also articulated category effects in lateral occipital cortex (Weiner et al. 2010), and
category specificity in those lateral occipital regions has been dissociated using TMS
(Pitcher et al. 2009).
The organization by semantic category in the ventral object-processing stream
described above is largely invariant to the task and stimuli used in the experiment (e.g.,
linguistic, image, auditory), although the responses are strongly modulated by task
and attention (Chao et al. 1999; Kanwisher and Downing 1998). In other words, what
determines the location of category-specific responses is the category (i.e., content)
of the stimulus and not its format. Category-specific responses in the ventral stream
are also generally invariant to stimulus manipulations such as orientation, size, and
contrast (Avidan et al. 2002; Levy et al. 2001; see figure 4.2, plate 6, for some examples
of category specificity in the ventral stream). Importantly, what seems to matter for
driving category-specific responses in the ventral stream is not so much the physical
stimulus, but the interpretation that is applied to a stimulus. For instance, simple geo-
metric shapes that move in either an animate or mechanical way, drive neural activity
in the tool-specific and animal-specific brain networks, respectively (e.g., Martin and
Weisberg 2003).
There is general agreement that the format of information represented in temporo-
occipital regions exhibiting category-specific responses is something like high-level
visual representations. Damage to the fusiform gyrus or lingual gyrus is known to
produce various types of visual agnosia, including color agnosia, and sometimes
alexia when the damage is restricted to the left hemisphere (Miceli et al. 2001; Stasenko
et al., 2014) or prosopagnosia when damage involves the right hemisphere. Similarly,
damage to lateral occipital cortex can lead to profound visual-form agnosia, as in the
very well-studied patient DF (Goodale et al. 1991).

4.6.3 Category Specificity in the Dorsal Object-Processing Stream

Tools, compared with a range of baseline stimuli, differentially drive BOLD contrast in
the left posterior middle temporal gyrus, left parietal cortex, and left premotor cortex
(figure 4.2, plate 6). The left middle temporal region that exhibits differential BOLD
responses when viewing manipulable objects (e.g., Martin et al. 1996; Thompson-
Schill et al. 1999; for a review, see Devlin et al. 2002) plays an important role in
processing the semantics of actions (e.g., Kable, Lease-Spellmeyer, and Chatterjee 2002;
Kemmerer et al. 2008; Martin et al. 1995) as well as mechanical (i.e., unarticulated)
The Representation and Organization of Object Concepts 93


1 Left ventral premotor cortex

2 Left intraparietal sulcus
3 Left medial fusiform gyrus
4 Right lateral fusiform gyrus
5 Middle temporal gyrus

p = 10–6 0.05 0.05 10–6

Figure 4.2 (plate 6)

Category-specific patterns of BOLD response in the healthy brain. This figure shows in red a
network of regions that are differentially activated for living animate things, and in blue, a
network of regions that are differentially activated for nonliving things (Data from Chao et al
2002; figure reproduced from Martin and Chao, 2001, with permission).

motion (Beauchamp et al. 2002, 2003; Martin and Weisberg 2003; Wheatley, Milleville,
and Martin 2007).
Regions of bilateral dorsal occipital cortex, posterior parietal cortex, through to the
anterior intraparietal sulcus, are automatically activated when participants observe
manipulable objects (e.g., Chao and Martin 2000; Fang and He 2005). Dorsal occipital
and posterior parietal regions are important for determining volumetric and spatial
information about objects relevant to pointing, while the anterior intraparietal sulcus
is thought to be important for hand preshaping for object prehension (Binkofski et al.
1998; Culham et al. 2003; Frey et al. 2005). Optic ataxia, an impairment in reaching
or grasping objects, is classically associated with lesions to posterior and superior pari-
etal structures (e.g., Pisella et al. 2000). Optic ataxia is not necessarily associated with
94 Bradford Z. Mahon

difficulties in manipulating objects according to their function (as optic ataxia patients
may be able to manipulate the objects once they are in hand).
Viewing or naming tools also differentially activates the left inferior parietal lobule
(e.g., Mahon et al. 2007; Rumiati et al. 2004), a region that is important for representing
complex object-associated manipulations. This region is also activated when congeni-
tally blind individuals think about tools, indicating that visual experience with objects
is not necessary for the specificity to emerge (Mahon, Schwarzbach, and Caramazza
2010). Damage to the left inferior parietal lobule is classically associated with apraxia
of object use (Rothi, Ochipa, and Heilman 1991; Rushworth, Krams, and Passingham
1997; Johnson-Frey 2004; see also discussion above in the context of the embodied
cognition hypothesis).

4.6.4 Implications of the Imaging Evidence

The power of the functional MRI approach to studying category specificity is that it
provides a window into all regions that are involved in processing information about
different categories, regardless of whether involvement of those regions is necessary.
Several different accounts of the causes of category-specific neural responses in humans
have been suggested. Most of those accounts are directed at understanding the causes
of category specificity in the ventral stream (Grill-Spector and Malach 2004; Haxby
et al. 2001; Martin 2007; Mahon et al. 2007; Mechelli et al. 2006; Rogers et al. 2005). It
is generally agreed that differential BOLD responses for tools in dorsal stream regions
are driven by automatic extraction of motor-relevant information (e.g., Martin 2007;
see discussion above in the context of the embodied cognition hypothesis).
Our own view of the causes of category specificity in the ventral stream is that it
emerges because different regions of the ventral stream are innately connected with
other regions of the brain that process nonvisual information about the same catego-
ries (for discussion of this connectivity-constrained account of category specificity, see
Mahon et al. 2007, 2009; Mahon and Caramazza 2009, 2011). The core aspect of this
proposal is that connectivity is what is innate and what drives domain specificity. In
other words, the domain specificity of a given region is not driven (only) by organiza-
tional principles expressed over information local to that region, but by the broader
network architecture in which that region is embedded (Mahon and Caramazza 2011).
Thus, for instance, the regions that exhibit specificity for tools (medial fusiform gyrus)
do so because that region has (by hypothesis) privileged connectivity with motor-
relevant structures that are involved in actually manipulating objects (Mahon et al
2007). Thus, the high-level visual representations for tools occupy those regions of the
visual system that are already connected with other regions of the brain that process
motor-relevant information about tools. Similarly, the argument would be that faces
are represented in regions of high-level visual cortex that have privileged connectivity
to regions of the brain that process affective information. By hypothesis, regions of the
The Representation and Organization of Object Concepts 95

lateral occipital cortex that differentially respond to images of the hands will express
privileged connectivity to somatomotor areas that also represent the hands (Bracci
et al. 2012). This kind of a connectivity-constrained account (Riesenhuber 2007) can
explain why there would be specialization for printed words, a class of stimuli for which
there can be no plausible evolutionary history (see Martin 2006). In other words, the
fact that there is specialization for printed words in the same way that there is special-
ization for categories that could have evolutionarily significant histories (faces, tools,
places, etc.), suggests that what is innate is not the content of the category, but rather
a basic scaffolding of connectivity between high-level visual regions and other parts
of the brain. Because those other parts of the brain will have their own biases toward
specifics functions, innately specified connectivity would be sufficient to drive spe-
cialization by semantic category in high-level visual regions. In the case of the visual
word-form area in the ventral object-processing stream, the prediction is made that it
will exhibit privileged connectivity to left hemisphere language regions (see also Plaut
and Behrmann 2011; for data from the domain of faces, see Thomas et al. 2009).
Several recently reported lines of evidence support the view that semantic domain
innately constrains the organization of object knowledge, and that connectivity is the
substrate for domain specificity. There are three strands of this evidence.

1. There are deep similarities between monkeys and humans in the “semantic space” of object
representations in the ventral stream. An expectation of the view that innate constraints
shape category specificity in the ventral stream is that such specificity, at least for some
categories, will also be found in nonhuman primates. It is well known, using neuro-
physiological recordings, that preferences for natural object stimuli exist in the inferior
temporal (IT) cortex of monkeys (e.g., Kiani et al. 2007; Tanaka et al. 1991), comparable
to observations with similar methods in awake human subjects (Kreiman, Koch, and
Fried 2000). More recently, functional imaging with macaques (Tsao et al. 2006) and
chimpanzees (Parr et al. 2009) suggests that at least for the category of faces, clusters of
face-preferring voxels can be found in the temporal cortex in monkeys, comparable to
what is observed in humans. Such common patterns of neural organization for some
classes of items in monkeys and humans could, of course, be entirely driven by dimen-
sions of visual similarity, which are known to modulate responses in the IT cortex
(Op de Beeck, Wagemans, and Vogels 2001). Even when serious attempts have been
made to explain such responses in terms of dimensions of visual similarity, however,
taxonomic structure emerges over and above the contribution of known visual dimen-
sions. For instance, Kriegeskorte and colleagues (2008) used multivoxel pattern analysis
to compare the similarity structure of a large array of different body, face, animal, plant,
and artifact stimuli in the monkey IT cortex and the human temporo-occipital cortex.
The similarity among the stimuli was measured in terms of the similarity in patterns
of brain responses they elicited, separately on the basis of the neurophysiological data
96 Bradford Z. Mahon

(monkeys, Kiani et al. 2007) and fMRI data (humans). The similarity structure that
emerged revealed a tight taxonomic structure common to monkeys and humans (see
figure 4.3, plate 7). Importantly, that similarity structure was not present in early visual
cortex and could not be reproduced from computational models of low-level visual
processing (see Kriegeskorte et al. 2008 for details and discussion).
2. There is an innate component to face recognition. Two recent reports highlight greater
neural or functional similarity between monozygotic twin pairs than between dizy-
gotic twin pairs (for discussion, see Park, Newman, and Polk 2009; Zhu et al. 2010).
The strength of these studies is that experiential contributions are held constant across
the two types of twin pairs. In an fMRI study, Polk and colleagues (2007) studied the
similarity between twin pairs in the distribution of responses to faces, houses, pseudo-
words, and chairs in the ventral stream. The authors found that face and place-related
responses within face and place-selective regions, respectively, were significantly
more similar for monozygotic than for dizygotic twins. In another study, Wilmer and
colleagues (2010) studied the face recognition and memory abilities in monozygotic
and dizygotic twin pairs (using some of the tests developed by Duchaine and Nakayama
2006). Wilmer and colleagues found that the correlation in performance on the face-
recognition task for monozygotic twins was more than double that for dizygotic twins.
This difference was not present for control tasks of verbal and visual memory, indicat-
ing selectivity in the genetic contribution to facial recognition abilities (see also Zhu
et al. 2010).
3. Category-specific neural organization does not require visual experience. Recent findings
indicate that visual experience is not necessary for the same, or similar, patterns of
category specificity to be present in the ventral stream. In an early positron emission
tomography study, Büchel and colleagues (1998) showed that congenitally blind sub-
jects show activation for words (presented in Braille) in the same region of the ventral
stream as sighted individuals (presented visually; see also Reich et al. 2011). Pietrini
and colleagues (2004) used multivoxel pattern analysis to show that the pattern of acti-
vation over voxels in the ventral stream was more consistent across different exemplars
within a category than exemplars across categories. More recently, we have shown that
the same medial-to-lateral bias in category preferences on the ventral surface of the
temporo-occipital cortex that is present in sighted individuals is present in congeni-
tally blind subjects (Mahon et al. 2009). Specifically, in congenitally blind participants,
nonliving things, compared to animals, elicited stronger activation in medial regions
of the ventral stream (see figure 4.4, plate 8).

Although these studies on category specificity in blind individuals represent only

a first-pass analysis of the role of visual experience in driving category specificity in
the ventral stream, they indicate that visual experience is not necessary for category
specificity to emerge in the ventral stream. Although this is not incompatible with


Figure 4.3 (plate 7)

Dendrograms showing similarity of response patterns across visual stimuli in monkey inferior
temporal (IT) cortex and human ventral temporal cortex. Kriegeskorte and colleagues (2008) ana-
lyzed neurophysiological data from monkey IT cortex and human fMRI data when participants
(monkeys, humans) were viewing numerous stimuli from many different categories. The similar-
ity of the neural responses across the stimuli was analyzed separately for monkeys and humans.
The figures, reproduced from Kriegeskorte and colleagues (2008, figure 4.4) use hierarchical clus-
tering to describe the similarity space of the stimuli. The fascinating aspect of these data is that
they show, with entirely independent analysis pipelines, that high-level visual cortex in monkeys
and humans represents largely the same similarity space for visual stimuli. Figure reproduced with
permission, from Kriegeskorte and colleagues (2008).
Sighted: Sighted: picture viewing
picture viewing Left ventral ROI Right ventral ROI
2 2

t Values (living - nonliving)

1 1
0 0
-1 -1
-2 -2
-3 -3
-4 -4
-40 -38 -36 -34 -32 -30 -28 -26 -24 24 26 28 30 32 34 36 38 40
Tal. Coord. X Dim Tal. Coord. X Dim

Sighted: Sighted: auditory task

auditory task Left ventral ROI Right ventral ROI
1 1
t Values (living - nonliving)

0.5 0.5

0 0

-0.5 -0.5

-1 -1

-1.5 -1.5
-40 -38 -36 -34 -32 -30 -28 -26 -24 24 26 28 30 32 34 36 38 40
Tal. Coord. X Dim Tal. Coord. X Dim

Congenitally blind: Congenitally blind: auditory task

auditory task Left ventral ROI Right ventral ROI
0 0
t Values (living - nonliving)

-0.5 -0.5
-1 -1
-1.5 -1.5
-2 -2
-2.5 -2.5

-40 -38 -36 -34 -32 -30 -28 -26 -24 24 26 28 30 32 34 36 38 40

Tal. Coord. X Dim Tal. Coord. X Dim

Figure 4.4 (plate 8)

Category-specific organization does not require visual experience. Congenitally blind and sighted
participants were presented with spoken words of living things (animals) and nonliving things
(tools, nonmanipulable objects) and were asked to make size judgments about the referents of
the words. The sighted participants were also shown pictures corresponding to the same stim-
uli in a separate scan. For sighted participants viewing pictures, the known finding was repli-
cated that nonliving things such as tools and large nonmanipulable objects lead to differential
neural responses in medial aspects of ventral temporo-occipital cortex. This pattern of differen-
tial BOLD responses for nonliving things in medial aspects of ventral temporo-occipital cortex
was also observed in congenitally blind participants and sighted participants performing the size
judgment task over auditory stimuli. These data indicate that the medial-to-lateral bias in the
distribution of category-specific responses does not depend on visual experience. For details of the
study, see Mahon and colleagues (2009). Figure reproduced from Mahon and colleagues (2009)
with permission.
The Representation and Organization of Object Concepts 99

the view that visual experience has an important role to play in shaping the organiza-
tion of high-level visual areas in sighted individuals, it does point to an organizational
constraint that cannot be reduced, in its entirety, to visual experience. The hypoth-
esis that we have advanced (Mahon and Caramazza 2011), and which I believe the
functional imaging data from blind individuals support, is that endogenously speci-
fied connectivity is the basic organizational constraint, or scaffolding, within which
experience-dependent organizational principles operate. Furthermore, and by hypoth-
esis, that endogenously specified connectivity will have a granularity that matches
the domains for which neural specificity has been described in the ventral object-
processing pathway.

4.7 Current and New Directions

Based on the brief overview of existing theories and evidence summarized above,
several conclusions can be extrapolated, which I pull together here. I also try to frame
what I believe are the issues that will drive research as the field moves forward, and
beyond the issues with which it has been occupied over the last several decades. My
approach to these prospective suggestions is to outline forward-leaning conclusions
that indicate new questions. The broad and overarching suggestion is that there will be
a common answer to the set of questions that are outlined. In short form, that answer
is that the only way to gain a deeper understanding of the informational content,
the organization of that content, and the dynamics of the conceptual system and its
interface with other systems will be through research that unpacks the structure and
function of connectivity.

4.7.1 Empirical Generalization I: Category-Specific Phenomena Cannot

Be Dissolved
As described above, very rich phenomena of category specificity in the human brain
were initially discovered by Warrington and her collaborators in brain-damaged
patients in the eighties, and with functional imaging in humans by Martin, Allison,
McCarthy, Kanwisher, and others in the nineties. Since those initial discoveries, much
of the research on category specificity in the human brain has been concerned with
characterizing the boundaries of the phenomena. In the context of patients with
category-specific semantic deficits, emphasis has been placed on whether categories
vary along continuous dimensions, such as familiarity, lexical frequency, structural
or visual complexity, distinctiveness of their critical features, relevance (and so on).
The common theoretical supposition behind those approaches is that category-specific
deficits arise as a result of difficulties with one (or multiple) dimensions that are corre-
lated with a semantic category boundary. In the context of category-specific responses
in the ventral object-processing stream, there has been an emphasis on understanding
100 Bradford Z. Mahon

the visual dimensions that putatively drive an organization by category in the

ventral stream. The common theoretical assumption is that category specificity in the
ventral stream arises through a type of experience-dependent coagulative process by
which dimensions of organization native to the visual system combine in either linear
or nonlinear ways to result in a “lumpy” organization by category.
Because of the common theoretical suppositions that have driven research on
category specificity in the brain, much experimental work has focused on parametri-
cally manipulating a dimension of interest and studying how category-specific phe-
nomena are modulated as a function of that dimension. This approach has been
enormously important for describing the boundaries of category-specific phenom-
ena. But perhaps the most important outcome of all this research is that we still have
category-specific phenomena that need explanation. In other words, a dimension
could have been discovered that when controlled or parametrically varied would have
“absorbed” the category effects into the dimension. That has not been the case—we
are left with the conclusion that category-specific phenomena are insoluble into the
continuous dimensions that have been identified to vary by category. Although there is
still much ongoing work that will flesh out the details of this conclusion, my prospec-
tive suggestions presuppose that this conclusion will endure. This then frames anew
an old question:

Question I: What neural and cognitive constraints drive an organization of object

knowledge by semantic category in the human brain?

4.7.2 Empirical Generalization II: Important Aspects of Conceptual Processing

Are Not Embodied
Research on the putative embodiment of concepts and the role of the motor system in
supporting perceptual analysis of actions has focused on demonstrations that the motor
system is activated across a range of situations that would not seem to necessitate motor
activation. There are multiple ways in which motor activation during conceptual and
perceptual analysis can be interpreted, ranging from the view that motor activation
constitutes a necessary and intermediary step in conceptual and perceptual analysis, to
the claim that motor activation is entirely ancillary to, and irrelevant for, conceptual
and perceptual analysis. The fact that multiple patient studies have reported associa-
tions between motor impairments and conceptual or perceptual impairments would
seem to rule out the view that motor activation is entirely irrelevant for conceptual
and perceptual analysis. On the other hand, dissociative patient studies demonstrating
that motor abilities can be compromised while sparing conceptual or perceptual abili-
ties rule out the view that motor activation is necessary or constitutive of conceptual
or perceptual analysis.
I think that the situation here is analogous to asking whether the function of a car
engine is embodied in the movement of the car—it depends on what you mean. If the
The Representation and Organization of Object Concepts 101

car is in gear, then there will be a direct mapping of turns of the tires on the road to
revolutions of the crank in the engine. But the car can be put in neutral, in which case
the engine can turn independently of the wheels. In fact, it is precisely this property
that makes gears (i.e., the transmission) such a useful interface between the engine and
the tires—gears are what give a car the flexibility to start at a dead stop and go to maxi-
mal speed using a single engine. Thus, asking why motor activation attends conceptual
processing is like asking why the wheels turn when the engine turns—the answer is
because the car is in gear and the engine is connected with the wheels via the transmis-
sion. The answer is the same if you ask why the engine turns if you push a car that is in
gear (i.e., jump starting; see, e.g., Glenberg, Sato, and Cattaneo 2008).
A counterpoint to this analogy is that cars would not be of any use if they did
not move—that is, a car that only sat in neutral would not really “be” a car. So there
is a priority placed on being in gear, but at the same time, there is not a one-to-one
relationship between the output of the motor while in gear and the turning of the
tires—the force given off by the engine must be interpreted into a format that can
be implemented into turns of the wheels. Likewise, conceptual processing, with no
interface with the world, would not be particularly useful. But this objection somewhat
misplaces the original question: the theoretical issue at stake concerns the format of
conceptual processing, and whether it is dissociable from sensory and motor knowl-
edge. And in that regard, the analogy is robust, in that even though the utility of cars
is expressed when the engine is engaged with the wheels, turns of the engine and turns
of the wheels are dissociable (cf. being in neutral). So although it may make sense at
one level to think of the function of an engine as being “embodied” in the movement
of the car—that is, at the level of understanding a particular state the car can be in—the
more basic point is that the function of the engine does not depend, in any constitu-
tive, logical, or necessary way, on the wheels.
Imagine that our state of knowledge of the function of cars was what it is with regard
to the human brain, and one observed that turns of the engine were related to turns of
the tires. One might then reasonably ask, à la embodied cognition, whether the engine
and the tires were really one and the same process. This is where I would suggest we are
with respect to understanding the relationship between sensorimotor activation and
conceptual processing. But once one discovered the behavior of the car when it was in
neutral, then the question would logically shift to asking how the engine was connected
to the tires. Similarly, I believe that the key issue we face as a field with respect to issues
of embodiment does not have to do with demonstrating that cognition can behave in
a way that indicates it can be synchronous with, or sensitive to, sensorimotor process-
ing; the key open issue concerns how to understand the structure and dynamics of the
interface between concepts and the sensorimotor systems.

Question II: If motor processes are not necessary for conceptual or perceptual processes,
then why are they automatically engaged during conceptual and perceptual processing?
102 Bradford Z. Mahon

4.7.3 A Consilience for Questions I and II

The suggestion of this chapter is that Questions I and II have a common answer, as yet
unspecified in its details, but which in broad strokes consists of a theory of connectiv-
ity among sensorimotor and conceptual representations. The strategy within the field
has been to divide and then reconnect. In that context, it is generally understood that
connectivity does not itself constitute information—the information is represented in
the local regions that are connected, and the connections are something like passive
conduits that pass bundles of information from region to region. The suggestion here is
that connectivity is itself a computation that underlies conceptual analysis. This would
require that we consider the performance of the entire network of regions as a unit of
analysis, and regard the information represented by the function of the whole network
as (at least at one level) a unit of analysis.
According to the task in which participants are engaged, stimuli will be analyzed
by the same network in different ways, and information distributed throughout the
network will be combined in different orders, with different weights, and to different
ends. As an example, consider how the dorsal and ventral visual pathways interact
when you reach out to pick up a hammer in order to simply move it over six inches
versus picking it up with the intention of using it to hammer a nail. When you pick
it up to move it, the grasp point on the object need only be calibrated such that the
object is picked up efficiently (i.e., at its center of mass) and such that the grasp does
not preclude, either because of the arrangement of other objects in the world or for bio-
mechanical constraints, the (planned or intended) final position of the hammer once it
is put down. When you pick up a hammer in order to use it, however, you explicitly do
not pick it up at its center of mass, precisely to capitalize on the lever-like properties of
hammer manipulation that allow the function of the object to be realized. The systems
that recognize the object as such, and interface with systems that represent the behav-
ioral intentions, are dissociable from the visuo-motor analysis that actually gets the
hand from its starting position to its grip point on the object. Thus, while the same set
of regions may be activated both when picking up a hammer to move it and when pick-
ing up a hammer to use it, the role that the different types of knowledge we have about
hammers play in shaping the overall behavior is very different. What distinguishes the
two situations is not therefore the types of information that are accessed (at some point
in the action) but the order in which those different types of information are accessed
and the weight that the different types of information are given in shaping the overall
behavior (for discussion, see Mahon and Wu, forthcoming). Such dynamic reordering
and reweighting of information must be mediated by connectivity, because the dif-
ferent types of information are known to be represented by dissociable brain regions.
Decisions that are made by our cognitive systems about how to act on the world
(e.g., what to attend to, how to shape the hand to grasp an object, that an object
should be grasped) are not made in isolation: such decisions are made in concert with
The Representation and Organization of Object Concepts 103

information that is computed about the world, including the current state of our body,
as well as our behavioral goals and internal states, by widely distributed and dissocia-
ble neurocognitive systems. Perception is not just for its own sake; perception always
occurs in the context of a behavioral goal, whether that goal is to take a drink of coffee
or simply to look at or inspect the mug. The suggestion here is that the structure and
dynamics of the conceptual system can be understood as a result of the varied pressures
made on the conceptual system in the service of different behavioral goals, or tasks,
and the consequent need to integrate qualitatively different types of information.
Those pressures can be understood along multiple time scales, including phylogenetic,
ontogenetic, and online-processing time scales.
The structure of neurocognitive information models is typically constrained by an
analysis of what information is necessary to complete a given task. However, more than
what is “strictly necessary” to perform a task, conceptual processing is also sensitive to
information that is available but may not be (strictly speaking) necessary to perform
the task. This notion of the availability of multiple types of information that could
bear on a given cognitive decision is what we have referred to as cognitive promiscuity
(for discussion, see Mahon and Caramazza 2008). Cognitive promiscuity, implemented
through the dynamics of information exchange among sensory, motor, and conceptual
representations, is the foothold for understanding why the motor system is activated
during conceptual and perceptual analysis. Cognitive promiscuity is also responsible,
by hypothesis, for the need to integrate high-level visual analysis with motor-relevant
computations about manipulable objects, or computations about the affective quality
of facial expressions with visual information about faces, or phonology with repre-
sentations of printed words; thus, cognitive promiscuity is the umbrella property that
motivates why connectivity between regions of the ventral stream and other regions
of the brain would drive specialization by semantic category in the ventral stream.
Understanding the constraints that shape the organization of the conceptual system
then becomes a project of unpacking the dynamics of cognitive promiscuity.

4.7.4 One Answer for a Distributed Question

It is a well-established fact that conceptual information is widely distributed in the
brain, in the sense that different aspects, or parts, of a concept are represented in disso-
ciable brain systems. For instance, for the concept HAMMER, knowledge about its typical
visual appearance and structure, knowledge about the sounds created when hammer-
ing, knowledge about the function of hammers, and knowledge about how to actu-
ally use hammers are all represented in dissociable brain systems. To date, the general
approach in the field has been to dissect out the components of concepts, sometimes
referred to as the features of the concept, and describe the principles (cognitive and
neural) that affect the representation and organization of each individual part. But
that would be like treating leg pain with only an understanding of the function and
104 Bradford Z. Mahon

physiology of the leg, and no understanding of how compression of spinal nerve fibers
might (remotely) cause leg pain. In order to understand the constraints that shape
specificity over one type of information (e.g., visual) in one part of the brain for a
given category of items, it is critical to understand how the visual information about
that class of items is integrated with nonvisual information about the same category
represented by other brain regions.
Although functional imaging or anatomical studies of white matter tractography
might seem to be the most obvious means for understanding connectivity in humans,
they are by no means the only approach. In particular, many rich empirical phenom-
ena have been taken as motivation for hypotheses about the format of conceptual
representations (i.e., that they are embodied) that may be better understood as phe-
nomena that result from the connectivity of the system. For instance, as noted above,
making repetitive arm movements, such as moving beans from a close to a far con-
tainer, can lead to slower responses for judging the grammaticality of sentences that
describe actions away from the body (Glenberg, Sato, and Cattaneo 2008). This has
been taken to mean that the understanding of the sentence “You passed the salt to
your friend’” involves, as a constitutive part, motor simulation. An alternative explana-
tion is that understanding that sentence occurs without intervening access to motor
information, but that the state of the motor system is not irrelevant for understanding
the sentence. In other words, the decision mechanism that oversees the grammatical-
ity judgment may be sensitive to information that is not part of the grammaticality
judgment, but that being available cannot be ignored, and it therefore affects response
time. Thus, rather than asking whether the format of lexical semantic representations
is motoric, it may be more productive to ask about the nature of the decision mecha-
nism involved in making judgments about sentences, and the types of information to
which that decision mechanism is sensitive.
Another class of data that could be brought to bear on understanding the cognitive
promiscuity of the conceptual system are the patterns of association and dissociation
of function observed in brain-damaged patients. Findings from brain damaged patients
are generally emphasized in the measure to which different types of information dis-
sociate from one another. Such dissociations are critical for drawing inferences about
the functional independence of different types of knowledge. However, of particular
importance for informing a theory of connectivity may be associations of impairments.
Price, Friston, and colleagues (Friston and Price 2011; Price and Friston 2002; Price
et al. 2001) have explored what they refer to as dynamic diaschisis. Dynamic diaschisis
is the idea that damage to one region of the brain can alter, and potentially impair,
the function of anatomically remote but functionally interconnected regions. For
instance, consider the fact that all types of knowledge (visual-perceptual and functional-
associative) are impaired for the damaged categories in patients with category-specific
deficits (figure 4.1B, plate 5). Such patterns of associated impairments are ambiguous
The Representation and Organization of Object Concepts 105

between the brain damage actually destroying tissue that is critical for representing the
different types of knowledge (either the same region or neighboring regions) and the
damage propagating at a functional level through dynamic diaschisis.
Concepts, as they are deployed in the service of behavior, are more than the sum
of their parts. As functionally unified representations, they allow the flexible recom-
bination of information based on inferences that go beyond the dissociable pieces of
information that form the concept. Connectivity, as the basis for the functional integ-
rity of concepts, lies at the heart of how concepts are both distributed and function-
ally unified. Thus, my argument here has been that characterizing connectivity at a
cognitive and neural level is not just an incremental step that will finally allow us to
understand how the different parts of the system are wired together and how informa-
tion is communicated among regions. Connectivity is an information-bearing property
of the system that must be understood on its own terms. Furthermore, I would suggest
that connectivity is the key to unlocking the reason why there is neural specificity for
different categories in the first place, and why the motor system is engaged in many
tasks that we know can be completed without motor information.


I would like to thank Eric Margolis and Stephen Laurence, as well as Alena Stasenko,
Frank Garcea, and Jessica Cantlon, for their comments on an earlier version of this
manuscript. I would like to thank Alfonso Caramazza for the many years of discussion
on these issues. I am grateful to Alex Martin and Niko Kriegeskorte for providing the
graphics in figures 4.2 and 4.3 (plates 6 and 7), respectively. Preparation of this manu-
script was supported in part by NIH grant R21 NS076176.


Allison, T., G. McCarthy, A. Nobre, A. Puce, and A. Belger. 1994. Human extrastriate visual cortex
and the perception of faces, words, numbers and colors. Cerebral Cortex 4:544–554.

Allport, D. A. 1985. Distributed memory, modular subsystems and dysphasia. In Current Perspec-
tives in Dysphasia, ed. S. K. Newman and R. Epstein, 207–244. New York: Churchill Livingstone.

Avidan, G., R. Hasson, T. Hendler, E. Zohary, and R. Malach. 2002. Analysis of the neuronal selec-
tivity underlying low fMRI signals. Current Biology 12:964–972.

Bar, M., and E. Aminoff. 2003. Cortical analysis of visual context. Neuron 38:347–358.

Barbarotto, R., M. Laiacona, V. Macchi, and E. Capitani. 2002. Picture-reality decisions, semantic
categories and gender: A new set of pictures, with norms and an experimental study. Neuropsycho-
logia 40:1637–1653.
106 Bradford Z. Mahon

Barsalou, L. W. 1999. Perceptual symbol systems. Behavioral and Brain Sciences 22:637–660.

Barsalou, L. W., W. K. Simmons, A. K. Barbey, and C. D. Wilson. 2003. Grounding conceptual

knowledge in modality-specific subsystems. Trends in Cognitive Sciences 7:84–91.

Beauchamp, M. S., K. E. Lee, J. V. Haxby, and A. Martin. 2002. Parallel visual motion processing
streams for manipulable objects and human movements. Neuron 34:149–159.

Beauchamp, M. S., K. E. Lee, J. V. Haxby, and A. Martin. 2003. Differential response to real and
point-light displays of moving humans and manipulable objects. Journal of Cognitive Neuroscience

Binder, J. R., and R. H. Desai. 2011. The neurobiology of semantic memory. Trends in Cognitive
Sciences 15:527–536.

Binkofski, F., C. Dohle, S. Posse, K. M. Stephan, H. Hefter, R. J. Seitz, and H. J. Freund. 1998.
Human anterior intraparietal area subserves prehension: A combined lesion and functional MRI
activation study. Neurology 50:1253–1259.

Bonner, M. F., and M. Grossman. 2012. Gray matter density of auditory association cortex relates
to knowledge of sound concepts in primary progressive aphasia. Journal of Neuroscience 32:

Bookheimer, S. Y. 2002. Functional MRI of language: New approaches to understanding the

cortical organization of semantic processing. Annual Review of Neuroscience 25:151–188.

Borgo, F., and T. Shallice. 2003. Category specificity and feature knowledge: Evidence from new
sensory-quality categories. Cognitive Neuropsychology 20:327–353.

Boulenger, V., L. Mechtouff, S. Thobois, E. Broussolle, M. Jeannerod, and T. A. Nazir. 2008. Word
processing in Parkinson’s disease is impaired for action verbs but not concrete nouns. Neuropsy-
chologia 46:743–756.

Boulenger, V., A. C. Roy, Y. Paulignan, V. Deprez, M. Jeannerod, and T. A. Nazir. 2006. Cross-talk
between language processes and overt motor behavior in the first 200 ms of processing. Journal of
Cognitive Neuroscience 18:1607–1615.

Bracci, S., C. Cavina-Pratesi, M. Ietswaart, A. Caramazza, and M. V. Peelen. 2012. Closely

overlapping responses to tools and hands in left lateral occipitotemporal cortex. Journal of Neuro-
physiology 107:1443–1456.

Büchel, C., C. Price, and K. Friston. 1998. A multimodal language region in the ventral visual
pathway. Nature 394:274–277.

Buxbaum, L. J., K. M. Kyle, and R. Menon. 2005. On beyond mirror neurons: Internal representa-
tions subserving imitation and recognition of skilled object-related actions in humans. Cognitive
Brain Research 25:226–239.

Buxbaum, L. J., and E. M. Saffran. 2002. Knowledge of object manipulation and object function:
Dissociations in apraxic and non apraxic subjects. Brain and Language 82:179–199.
The Representation and Organization of Object Concepts 107

Buxbaum, L. J., T. Veramonti, and M. F. Schwartz. 2000. Function and manipulation tool knowl-
edge in apraxia: Knowing “what for” but not “how.” Neurocase 6:83–97.

Capitani, E., M. Laiacona, B. Mahon, and A. Caramazza. 2003. What are the facts of category-
specific disorders? A critical review of the clinical evidence. Cognitive Neuropsychology 20:213–261.

Caramazza, A., and A. E. Hillis. 1991. Lexical organization of nouns and verbs in the brain. Nature

Caramazza, A., A. E. Hillis, B. C. Rapp, and C. Romani. 1990. The multiple semantics hypothesis:
Multiple confusions? Cognitive Neuropsychology 7:161–189.

Caramazza, A., and B. Z. Mahon. 2003. The organization of conceptual knowledge: The evidence
from category-specific semantic deficits. Trends in Cognitive Sciences 7:354–361.

Caramazza, A., and B. Z. Mahon. 2006. The organization of conceptual knowledge in the brain:
The future’s past and some future direction. Cognitive Neuropsychology 23:13–38.

Caramazza, A., and J. R. Shelton. 1998. Domain specific knowledge systems in the brain: The
animate-inanimate distinction. Journal of Cognitive Neuroscience 10:1–34.

Chao, L. L., J. V. Haxby, and A. Martin. 1999. Attribute-based neural substrates in temporal cortex
for perceiving and knowing about objects. Nature Neuroscience 2:913–919.

Chao, L. L., and A. Martin. 2000. Representation of manipulable manmade objects in the dorsal
stream. NeuroImage 12:478–484.

Chatterjee, A. 2010. Disembodying cognition. Language and Cognition 2:79–116.

Costa, A., A. Caramazza, and N. Sebastián-Gallés. 2000. The cognate facilitation effect: Implica-
tions for models of lexical access. Journal of Experimental Psychology: Learning, Memory, and
Cognition 26:1283–1296.

Cree, G. S., and K. MacRae. 2003. Analyzing the factors underlying the structure and computa-
tion of meaning of chipmunk, cherry, chisel, cheese, and cello (and many other such concrete
nouns). Journal of Experimental Psychology: General 132:163–201.

Crutch, S. J., and E. K. Warrington. 2003. The selective impairment of fruit and vegetable
knowledge: A multiple processing channels account of fine-grain category specificity. Cognitive
Neuropsychology 20:355–372.

Culham, J. C., S. L. Danckert, J. F. X. DeSourza, J. S. Gati, R. S. Menon, and M. A. Goodale. 2003.

Visually guided grasping produces fMRI activation in dorsal but not ventral stream brain areas.
Experimental Brain Research 153:180–189.

Dehaene, S., L. Cohen, M. Sigman, and F. Vinckier. 2005. The neural code for written words: A
proposal. Trends in Cognitive Sciences 9:335–341.

Devlin, J. T., R. P. Russell, M. H. Davis, C. J. Price, H. E. Moss, M. J. Fadili, and L. K. Tyler. 2002. Is
there an anatomical basis for category-specificity? Semantic memory studies in PET and fMRI.
Neuropsychologia 40:54–75.
108 Bradford Z. Mahon

Dinstein, I., C. Thomas, M. Behrmann, and D. J. Heeger. 2008. A mirror up to nature. Current
Biology 18:R13–R18.

di Pellegrino, G., L. Fadiga, L. Fogassi, V. Gallese, and G. Rizzolatti. 1992. Understanding motor
events: A neurophysiology study. Experimental Brain Research 91:176–180.

Downing, P. E., A. W. Chan, M. V. Peelen, C. M. Dodds, and N. Kanwisher. 2006. Domain speci-
ficity in visual cortex. Cerebral Cortex 16:1453–1461.

Duchaine, B., and K. Nakayama. 2006. The Cambridge face memory test: Results for neurologi-
cally intact individuals and an investigation of its validity using inverted face stimuli and prosop-
agnosic participants. Neuropsychologia 44:576–585.

Epstein, R., and N. Kanwisher. 1998. A cortical representation of the local visual environment.
Nature 392:598–601.

Fang, F., and S. He. 2005. Cortical responses to invisible objects in the human dorsal and ventral
pathways. Nature Neuroscience 8:1380–1385.

Farah, M. J., and C. Rabinowitz. 2003. Genetic and environmental influences on the organization
of semantic memory in the brain: Is “living things” an innate category? Cognitive Neuropsychology

Frey, S. H., D. Vinton, R. Newman-Norlund, and S. T. Grafton. 2005. Cortical topography

of human anterior intraparietal cortex active during visually-guided grasping. Cognitive Brain
Research 23:397–405.

Friston, K. J., and C. J. Price. 2011. Modules and brain mapping. Cognitive Neuropsychology

Funnell, E., and J. Sheridan. 1992. Categories of knowledge? Unfamiliar aspects of living and
nonliving things. Cognitive Neuropsychology 9:135–153.

Gallese, V., L. Fadiga, L. Fogassi, and G. Rizzolatti. 1996. Action recognition in the premotor
cortex. Brain 119:593–609.

Gallese, V., and G. Lakoff. 2005. The brain’s concepts: The role of the sensory-motor system in
conceptual knowledge. Cognitive Neuropsychology 22:455–479.

Garcea, F. E., and B. Z. Mahon. 2012. What is in a tool concept? Dissociating manipulation
knowledge from function knowledge. Memory & Cognition 40:1303–1313.

Gerlach, C. 2007. A review of functional imaging studies on category specificity. Journal of Cogni-
tive Neuroscience 19:296–314.

Glenberg, A. M., M. Sato, and L. Cattaneo. 2008. Use-induced motor plasticity affects the process-
ing of abstract and concrete language. Current Biology 18:R1–R2.

Goodale, M. A., and A. D. Milner. 1992. Separate visual pathways for perception and action.
Trends in Neurosciences 15:20–25.
The Representation and Organization of Object Concepts 109

Goodale, M. A., A. D. Milner, L. S. Jakobsen, and D. P. Carey. 1991. A neurological dissociation

between perceiving objects and grasping them. Nature 349:154–156.

Grill-Spector, K., and R. Malach. 2004. The human visual cortex. Annual Review of Neuroscience

Hart, J., Jr., R. Anand, S. Zoccoli, M. Maguire, J. Gamino, G. Tillman, R. King, and M. A. Kraut.
2007. Neural substrates of semantic memory. Journal of the International Neuropsychological Society

Hauk, O., I. Johnsrude, and F. Pulvermüller. 2004. Somatotopic representation of action words in
human motor and premotor cortex. Neuron 41:301–307.

Haxby, J. V., M. I. Gobbini, M. L. Furey, A. Ishai, J. L. Schouten, and P. Pietrini. 2001. Distributed
and overlapping representations of faces and objects in ventral temporal cortex. Science

Hickok, G. 2009. Eight problems for the mirror neuron theory of action understanding in
monkeys and humans. Journal of Cognitive Neuroscience 21:1229–1243.

Hickok, G. 2010. The role of mirror neurons in speech perception and action word semantics.
Language and Cognitive Processes 25:749–776.

Hickok, G., M. Costanzo, R. Capasso, and G. Miceli. 2011. The role of Broca’s area in speech
perception: Evidence from aphasia revisited. Brain and Language 119:214–220.

Hickok, G., K. Okada, W. Barr, J. Pa, C. Rogalsky, K. Donnelly, L. Barde, and A. Grant. 2008.
Bilateral capacity for speech sound processing in auditory comprehension: Evidence from Wada
procedures. Brain and Language 107:179–184.

Humphreys, G. W., and E. M. E. Forde. 2001. Hierarchies, similarity, and interactivity in object
recognition: “Category-specific” neuropsychological deficits. Behavioral and Brain Sciences 24:

Johnson-Frey, S. H. 2004. The neural bases of complex tool use in humans. Trends in Cognitive
Sciences 8:71–78.

Kable, J. W., J. Lease-Spellmeyer, and A. Chatterjee. 2002. Neural substrates of action event
knowledge. Journal of Cognitive Neuroscience 14:795–805.

Kanwisher, N., and P. Downing. 1998. Separating the wheat from the chaff. Science 252:57–58.

Kanwisher, N., J. McDermott, and M. M. Chun. 1997. The fusiform face area: A module in human
extrastriate cortex specialized for face perception. Journal of Neuroscience 17:4302–4311.

Kemmerer, D. Forthcoming. The Cognitive Neuroscience of Language: An Introduction. New York:

Psychology Press.

Kemmerer, D., J. Gonzalez Castillo, T. Talavage, S. Patterson, and C. Wiley. 2008. Neuroanatomi-
cal distribution of five semantic components of verbs: Evidence from fMRI. Brain and Language
110 Bradford Z. Mahon

Kemmerer, D., D. Rudrauf, K. Manzel, and D. Tranel. 2012. Behavioral patterns and lesion sites
associated with impaired processing of lexical and conceptual knowledge of actions. Cortex

Kiani, R., H. Esteky, K. Mirpour, and K. Tanaka. 2007. Object category structure in response
patterns of neuronal population in monkey inferior temporal cortex. Journal of Neurophysiology

Kiefer, M., and F. Pulvermüller. 2012. Conceptual representations in mind and brain: Theoretical
developments, current evidence and future directions. Cortex 48:805–825.

Kreiman, G., C. Koch, and I. Fried. 2000. Category-specific visual responses of single neurons in
the human medial temporal lobe. Nature Neuroscience 3:946–953.

Kriegeskorte, N., M. Mur, D. A. Ruff, R. Kiani, J. Bodurka, H. Esteky, K. Tanaka, and P. A.

Bandettini. 2008. Matching categorical object representations in inferior temporal cortex of man
and monkey. Neuron 60:1126–1141.

Laws, K. 2005. “Illusions of normality”: A methodological critique of category-specific naming.

Cortex 41:842–851.

Levy, I., U. Hasson, G. Avidan, T. Hendler, and R. Malach. 2001. Center-periphery organization of
human object areas. Nature Neuroscience 4:533–539.

Lewis, J. 2006. Cortical networks related to human use of tools. Neuroscientist 12:211–231.

Lhermitte, F., and M.-F. Beauvois. 1973. A visual speech disconnection syndrome: Report of a
case with optic aphasia, agnosic alexia and color agnosia. Brain 96:695–714.

Liberman, A. M., F. S. Cooper, D. P. Shankweiler, and M. Studdert-Kennedy. 1967. Perception of

the speech code. Psychological Review 74:431–461.

Liberman, A. M., and I. G. Mattingly. 1985. The motor theory of speech perception revised.
Cognition 21:1–36.

Lingnau, A., B. Gesierich, and A. Caramazza. 2009. Asymmetric fMRI adaptation reveals no
evidence for mirror neurons in humans. Proceedings of the National Academy of Sciences of the
United States of America 106:9925–9930.

Mahon, B. Z. 2008. Action recognition: Is it a motor process? Current Biology 18:R1068–R1069.

Mahon, B. Z., S. Anzellotti, J. Schwarzbach, M. Zampini, and A. Caramazza. 2009. Category-

specific organization in the human brain does not require visual experience. Neuron 63:397–405.

Mahon, B. Z., and A. Caramazza. 2003. Constraining questions about the organization and repre-
sentation of conceptual knowledge. Cognitive Neuropsychology 20:433–450.

Mahon, B. Z., and A. Caramazza. 2005. The orchestration of the sensory-motor system: Clues
from neuropsychology. Cognitive Neuropsychology 22:480–494.
The Representation and Organization of Object Concepts 111

Mahon, B. Z., and A. Caramazza. 2008. A critical look at the embodied cognition hypothesis and
a new proposal for grounding conceptual content. Journal of Physiology, Paris 102:59–70.

Mahon, B. Z., and A. Caramazza. 2009. Concepts and categories: A cognitive neuropsychological
perspective. Annual Review of Psychology 60:27–51.

Mahon, B. Z., and A. Caramazza. 2011. What drives the organization of object knowledge in the
brain? Trends in Cognitive Sciences 15:97–103.

Mahon, B. Z., S. Milleville, G. A. L. Negri, R. I. Rumiati, A. Caramazza, and A. Martin. 2007.

Action-related properties of objects shape object representations in the ventral stream. Neuron

Mahon, B. Z., J. Schwarzbach, and A. Caramazza. 2010. The representation of tools in left parietal
cortex independent of visual experience. Psychological Science 21:764–771.

Mahon, B. Z., and W. Wu. Forthcoming. Does conceptual analysis penetrate the dorsal visual
pathway? To appear in Cognitive Penetration, ed. J. Zeimbekis and A. Raftopoulos.

Martin, A. 2006. Shades of Dejerine—Forging a causal link between the visual word form area and
reading. Neuron 50:173–190.

Martin, A. 2007. The representation of object concepts in the brain. Annual Review of Psychology

Martin, A., and L. L. Chao. 2001. Semantic memory and the brain: Structure and processes.
Current Opinion in Neurobiology 11:194–201.

Martin, A., J. V. Haxby, F. M. Lalonde, C. L. Wiggs, and L. G. Ungerleider. 1995. Discrete cortical
regions associated with knowledge of color and knowledge of action. Science 270:102–105.

Martin, A., L. G. Ungerleider, and J. V. Haxby. 2000. Category specificity and the brain: The
sensory/motor model of semantic representations of objects. In The New Cognitive Neurosciences.
2nd ed., ed. M. S. Gazzaniga, 1023–1036. Cambridge, MA: MIT Press.

Martin, A., and J. Weisberg. 2003. Neural foundations for understanding social and mechanical
concepts. Cognitive Neuropsychology 20:575–587.

Martin, A., C. L. Wiggs, L. G. Ungerleider, and J. V. Haxby. 1996. Neural correlates of category-
specific knowledge. Nature 379:649–652.

Mechelli, A., G. Sartori, P. Orlandi, and C. J. Price. 2006. Semantic relevance explains category
effects in medial fusiform gyri. NeuroImage 15:992–1002.

Merigan, W. H., and J. H. Maunsell. 1993. How parallel are the primate visual pathways? Annual
Review of Neuroscience 16:369–402.

Miceli, G., E. Fouch, R. Capasso, J. R. Shelton, F. Tamaiuolo, and A. Caramazza. 2001. The disso-
ciation of color from form and function knowledge. Nature Neuroscience 4:662–667.
112 Bradford Z. Mahon

Negri, G. A. L., R. I. Rumiati, A. Zadini, M. Ukmar, B. Z. Mahon, and A. Caramazza. 2007. What is
the role of motor simulation in action and object recognition? Evidence from apraxia. Cognitive
Neuropsychology 24:795–816.

Ochipa, C., L. J. G. Rothi, and K. M. Heilman. 1989. Ideational apraxia: A deficit in tool selection
and use. Annals of Neurology 25:190–193.

Op de Beeck, H. P., J. Haushofer, and N. G. Kanwisher. 2008. Interpreting fMRI data: Maps,
modules, and dimensions. Nature Reviews: Neuroscience 9:123–135.

Op de Beeck, H., J. Wagemans, and R. Vogels. 2001. Inferotemporal neurons represent

low-dimensional configurations of parameterized shapes. Nature Neuroscience 4:1244–1252.

Park, J., L. I. Newman, and T. A. Polk. 2009. Face processing: The interplay of nature and nurture.
Neuroscientist 15:445–449.

Parr, L. A., E. Hecht, S. K. Barks, T. M. Preuss, and J. R. Votaw. 2009. Face processing in the
chimpanzee brain. Current Biology 19:50–53.

Patterson, K., P. J. Nestor, and T. T. Rogers. 2007. Where do you know what you know? The
representation of semantic knowledge in the human brain. Nature Reviews: Neuroscience 8:

Pazzaglia, M., L. Pizzamiglio, E. Pes, and S. M. Aglioti. 2008. The sound of actions in apraxia.
Current Biology 18:1766–1772.

Pazzaglia, M., N. Smania, E. Corato, and S. M. Aglioti. 2008. Neural underpinnings of gesture
discrimination in patients with limb apraxia. Journal of Neuroscience 28:3030–3040.

Peterson, R. P., and P. Savoy. 1998. Lexical selection and phonological encoding during language
production: Evidence for cascading processing. Journal of Experimental Psychology: Learning,
Memory, and Cognition 24:539–557.

Pietrini, P., M. L. Furey, E. Ricciardi, M. I. Gobbini, W. H. Wu, L. Cohen, M. Guazzelli, and J. V.

Haxby. 2004. Beyond sensory images: Object-based representation in the human ventral path-
way. Proceedings of the National Academy of Sciences of the United States of America 101:5658–5663.

Pisella, L., B. F. Binkofski, K. Lasek, I. Toni, and Y. Rossetti. 2006. No double-dissociation between
optic ataxia and visual agnosia: Multiple substreams for multiple visuo-manual integrations.
Neuropsychologia 44:2734–2748.

Pisella, L., H. Gréa, C. Tilikete, A. Vighetto, M. Desmurget, G. Rode, D. Boisson, and Y. Rossetti.
2000. An “automatic pilot” for the hand in human posterior parietal cortex: Toward reinterpret-
ing optic ataxia. Nature Neuroscience 3:729–736.

Pitcher, D., L. Charles, J. T. Devlin, V. Walsh, and B. Duchaine. 2009. Triple dissociation of faces,
bodies, and objects in extrastriate cortex. Current Biology 19:319–324.

Plaut, D. C., and M. Behrmann. 2011. Complementary neural representations for faces and
words: A computational perspective. Cognitive Neuropsychology 28:251–275.
The Representation and Organization of Object Concepts 113

Polk, T. A., J. Park, M. R. Smith, and D. C. Park. 2007. Nature versus nurture in ventral visual
cortex: A functional magnetic resonance imaging study of twins. Journal of Neuroscience 27:

Price, C. J., and K. J. Friston. 2002. Degeneracy and cognitive anatomy. Trends in Cognitive Sciences

Price, C. J., E. A. Warburton, C. J. Moore, R. S. Frackowiak, and K. J. Friston. 2001. Dynamic

diaschisis: Anatomically remote and context-sensitive human brain lesions. Journal of Cognitive
Neuroscience 13:419–429.

Pulvermüller, F. 2005. Brain mechanisms linking language and action. Nature Reviews: Neurosci-
ence 6:576–582.

Rapcsak, S. Z., C. Ochipa, K. C. Anderson, and H. Poizner. 1995. Progressive Ideomotor apraxia:
Evidence for a selective impairment of the action production system. Brain and Cognition 27:

Reich, L., M. Szwed, L. Cohen, and A. Amedi. 2011. A ventral visual stream reading center inde-
pendent of visual experience. Current Biology 21:363–368.

Riesenhuber, M. 2007. Appearance isn’t everything: News on object representation in cortex.

Neuron 55:341–344.

Rizzolatti, G., and L. Craighero. 2004. The mirror-neuron system. Annual Review of Neuroscience

Rogalsky, C., T. Love, D. Driscoll, S. W. Anderson, and G. Hickok. 2011. Are mirror neurons the
basis of speech perception? Evidence from five cases with damage to the purported human mirror
system. Neurocase 17:178–187.

Rogers, T. T., J. Hocking, A. Mechilli, K. Patterson, and C. Price. 2005. Fusiform activation
to animals is driven by the process, not the stimulus. Journal of Cognitive Neuroscience 17:

Rothi, L. J. G., C. Ochipa, and K. Heilman. 1991. A cognitive neuropsychological model of limb
praxis. Cognitive Neuropsychology 8:443–458.

Rumiati, R. I., P. H. Weiss, T. Shallice, G. Ottoboni, J. Noth, L. Zilles, and G. R. Fink. 2004. Neural
basis of pantomiming the use of visually presented objects. NeuroImage 21:1224–1231.

Rumiati, R. I., S. Zanini, L. Vorano, and T. Shallice. 2001. A form of ideational apraxia as a selec-
tive deficit of contention scheduling. Cognitive Neuropsychology 18:617–642.

Rushworth, M. F., M. Krams, and R. E. Passingham. 1997. The attentional role of the left parietal
cortex: The distinct lateralization and localization of motor attention in the human brain. Journal
of Cognitive Neuroscience 13:698–710.

Samson, D., and A. Pillon. 2003. A case of impaired knowledge for fruit and vegetables. Cognitive
Neuropsychology 20:373–400.
114 Bradford Z. Mahon

Sartori, G., and L. Lombardi. 2004. Semantic relevance and semantic disorders. Journal of
Cognitive Neuroscience 16:439–452.

Sartori, G., L. Lombardi, and L. Mattiuzzi. 2005. Semantic relevance best predicts normal and
abnormal name retrieval. Neuropsychologia 43:754–770.

Schenk, T. 2006. An allocentric rather than perceptual deficit in patient DF. Nature Neuroscience

Shmuelof, L., and E. Zohary. 2005. Dissociation between ventral and dorsal fMRI activation
during object action recognition. Neuron 47:457–470.

Simmons, W. K., and L. W. Barsalou. 2003. The similarity-in-topography principle: Reconciling

theories of conceptual deficits. Cognitive Neuropsychology 20:451–486.

Stasenko, A., F. E. Garcea, M. Dombovy, and B. Z. Mahon. 2014. When concepts lose their color:
A case of selective loss of knowledge of object color. Cortex 58: 217-238.

Stasenko, A., F. E. Garcea, and B. Z. Mahon. 2013. What happens to the motor theory of percep-
tion when the motor theory is damaged? Language and Cognition. 5: 225-238

Tanaka, K., H. Saito, Y. Fukada, and M. Moriya. 1991. Coding visual images of objects in the
inferotemporal cortex of the macaque monkey. Journal of Neurophysiology 66:170–189.

Thomas, C., G. Avidan, K. Humphreys, K. J. Jung, F. Gao, and M. Behrmann. 2009. Reduced
structural connectivity in ventral visual cortex in congenital prosopagnosia. Nature Neuroscience

Thompson-Schill, S. L. 2003. Neuroimaging studies of semantic memory: Inferring “how” from

“where.” Neuropsychologia 41:280–292.

Thompson-Schill, S. L., G. K. Aguirre, M. D’Esposito, and M. J. Farah. 1999. A neural basis for cat-
egory and modality specificity of semantic knowledge. Neuropsychology 37:671–676.

Toni, I., F. P. de Lange, M. L. Noordzij, and P. Hagoort. 2008. Language beyond action. Journal of
Physiology, Paris 102:71–79.

Tsao, D. Y., W. A. Freiwald, R. B. H. Tootell, and M. S. Livingstone. 2006. A cortical region consist-
ing entirely of face-selective cells. Science 311:669–674.

Tyler, L. K., H. E. Moss, M. R. Durrant-Peatfield, and J. P. Levy. 2000. Conceptual structure and
the structure of concepts: A distributed account of category-specific deficits. Brain and Language

Tyler, L. K., and H. E. Moss. 2001. Towards a distributed account of conceptual knowledge. Trends
in Cognitive Sciences 5:244–252.

Ungerleider, L. G., and M. Mishkin. 1982. Two cortical visual systems. In Analysis of Visual
Behavior, ed. D. J. Ingle, M. A. Goodale, and R. J. W. Mansfield, 549–586. Cambridge, MA: MIT
The Representation and Organization of Object Concepts 115

Vigliocco, G., D. P. Vinson, W. Lewis, and M. F. Garrett. 2004. Representing the meanings of
object and action words: The featural and unitary semantic space hypothesis. Cognitive Psychology

Vinson, D. P., G. Vigliocco, S. Cappa, and S. Siri. 2003. The breakdown of semantic knowledge:
Insights from a statistical model of meaning representation. Brain and Language 86:347–365.

Warrington, E. K., and R. A. McCarthy. 1983. Category specific access dysphasia. Brain

Warrington, E. K., and R. A. McCarthy. 1987. Categories of knowledge: Further fractionations

and an attempted integration. Brain 110:1273–1296.

Warrington, E. K., and T. Shallice. 1984. Category specific semantic impairments. Brain

Weiner, K. S., R. Sayres, J. Vinberg, and K. Grill-Spector. 2010. fMRI-Adaptation and category
selectivity in human ventral temporal cortex: Regional differences across time scales. Journal of
Neurophysiology 103:3349–3365.

Wheatley, T., S. C. Milleville, and A. Martin. 2007. Understanding animate agents: Distinct roles
for the social network and mirror system. Psychological Science 18:469–474.

Wilmer, J. B., L. Germine, C. F. Chabris, G. Chatterjee, M. Williams, E. Loken, K. Nakayama, and

B. Duchaine. 2010. Human face recognition ability is specific and highly heritable. Proceedings of
the National Academy of Sciences of the United States of America 16:5238–5241.

Zhu, Q., Y. Song, S. Hu, X. Li, M. Tian, Z. Zhen, Q. Dong, N. Kanwisher, and J. Liu. 2010. Herita-
bility of the specific cognitive ability of face perception. Current Biology 20:137–142.
5 Concept Nativism and Neural Plasticity

Stephen Laurence and Eric Margolis

5.1 Introduction1

One of the most important recent developments in the study of concepts has
been the resurgence of interest in nativist accounts of the human conceptual system.
However, many theorists suppose that a key feature of neural organization—the brain’s
plasticity—undermines the nativist approach to concept acquisition. In this chapter,
we argue that although this view of the matter has an initial air of plausibility, it gets
things exactly backward. Not only does the brain’s plasticity fail to undermine concept
nativism, but a detailed examination of the neurological evidence actually provides
powerful support for concept nativism, giving rise to what we call the argument from
neural wiring. The brain is most definitely an organ that is altered by experience, but as
we will see, the ways in which it is altered support a nativist perspective on cognitive

5.2 The Contemporary Empiricism-Nativism Debate

Nativist views of concepts trace back to Plato, Descartes, Leibniz, and other philoso-
phers who theorized about the existence of innate ideas. Partly for this reason, nativism
about the conceptual system is sometimes characterized simply as the view that there
are innate ideas or concepts, and empiricism as the view that the mind is initially a
blank slate in that it has no innate structure whatsoever. However, this way of distin-
guishing empiricism from nativism is ill-advised.
First, to characterize empiricism as the view that the mind begins with no innate
structure would have the unfortunate consequence of there not really being any empir-
icists. It has long been recognized by all parties to the empiricism-nativism debate
that a mind without any innate structure—a truly blank slate—wouldn’t be capable of
learning. There has to be something that accounts for why human beings come to know

1. This article was fully collaborative; the order of the authors’ names is arbitrary.
118 Stephen Laurence and Eric Margolis

anything at all about the world around them. As Quine once noted, even “the behav-
iorist is knowingly and cheerfully up to his neck in innate mechanisms” (Quine 1969,
95–96).2 Second, although it is true that nativists are more likely than empiricists to
embrace innate concepts in addition to other types of innate psychological structures,
focusing exclusively on whether concepts are innate doesn’t do justice to mainstream
views of the conceptual system. Empiricists may accept some innate concepts, and
nativists may hold that what matters is not innate concepts per se, but rather the exis-
tence of a rich innate basis for acquiring concepts.
For these reasons, we think it best to characterize concept nativism directly in terms
of what is at stake in the disagreement between empiricists and nativists. For contem-
porary theorists in philosophy and cognitive science, the disagreement revolves around
the character of the innate psychological structures that underlie concept acquisition.
According to empiricist approaches, there are few if any innate concepts, and con-
cept acquisition is, by and large, governed by a small number of innate general-purpose
cognitive systems being repeatedly engaged. Sometimes this point is put by saying
that empiricists claim that concepts are largely acquired on the basis of experience
and hence that the conceptual system is predominantly a product of learning. But
the crucial fact here isn’t that empiricists place a lot of weight on learning. (As we’ll
see in a moment, nativists do too.) Rather, what is characteristic of empiricism in
the empiricism-nativism debate is its distinctively empiricist approach to learning. The
empiricist view is that concept learning overwhelmingly traces back to general-purpose
cognitive systems and that these provide the psychological underpinning for the many
varied concepts that humans come to possess. For example, on a typical empiricist
view, concepts related to agency and concepts related to number are both the product
of the same kind of psychological processes embedded in the same general-purpose
concept-acquisition systems. The reason agency representations form in the one case
and numerical representations in the other is simply a reflection of the differing experi-
ences of the learner resulting in the different input to these systems.
The nativist approach, in contrast, holds that innate concepts and/or innate spe-
cial-purpose cognitive systems (of varying degrees of specialization) play a key role
in conceptual development, alongside general-purpose cognitive systems. So what is
characteristic of nativism in the empiricism-nativism debate is its distinctively nativist

2. Some contemporary theorists who undoubtedly fall on the empiricist side of the empiricism-
nativism divide have rejected the label empiricism because of its association with the view that the
mind lacks innate structure. For example, in a discussion relating work in neuroscience to theo-
ries of conceptual development, Steven Quartz remarks, “I have avoided using the term empiri-
cism, instead stating the strategy in terms of not being strongly innate. My reason for this lies in
the common identification of empiricism with Tabula Rasa learning” (Quartz 2003, 34). Since we
take there to be a substantive issue at stake between theorists like Quartz and concept nativists, a
better characterization of empiricism is clearly needed.
Concept Nativism and Neural Plasticity 119

approach to concept acquisition, crucially involving a substantial number of innate con-

cepts, numerous innate special-purpose systems involved in concept acquisition,
or, most likely, some combination of the two. A nativist view is perfectly at home
with the claim that representations of agency might depend on psychological pro-
cesses that reflect the operation of innate agency-specific concept-acquisition systems,
while representations of number depend on separate, innate number-specific concept-
acquisition systems. The reason agency representations form in the one case and
numerical representations in the other would then be due as much to the fact that they
are governed by different innate special-purpose acquisition systems as to the differing
input to these systems.
As frameworks for explaining concept acquisition, both nativism and empiricism
come in differing strengths. A strong form of empiricism would claim that there
are no innate concepts whatsoever and that concept acquisition depends exclusively
on a small number of general-purpose psychological systems. Jesse Prinz defends a
view along these lines, holding that concepts “are all learned, not innate” (Prinz 2005,
679). After arguing against what he takes to be the main proposals for special-purpose
innate concept-acquisition systems, he summarizes his discussion by noting, “I do
not believe that any of these domains is innate. That is to say, I do not think we have
innate domain-specific knowledge that contributes to structuring our concepts” (Prinz
2005, 688).
A weaker empiricist view might admit that there are a limited number of special-
purpose cognitive mechanisms that constrain how the conceptual system develops in
certain isolated cases, particularly cases supporting basic biological needs, but apart
from these few minor exceptions, concept acquisition is governed solely by general-
purpose cognitive systems. Rogers and McClelland (2004) defend such a view. After
arguing that a general-purpose connectionist model can explain how adult semantic
memory is organized, they suggest that there may be a handful of instances of prepared
learning, including, for example, a tendency to withdraw from strong stimuli and to
respond favorably to the taste of fat and sugar. Nonetheless, Rogers and McClelland
state that they are “reluctant … to accept that, in general, human semantic cognition
is prepared in this way,” arguing instead that “domain-general mechanisms can dis-
cover the sorts of domain-specific principles that are evident in the behavior of young
children” (Rogers and McClelland 2004, 369).
Nativist views also come in differing strengths. In fact, arguably one of the main rea-
sons concept nativism has had so few adherents is because of its association with one
of the most audacious nativist positions ever defended—Jerry Fodor’s radical concept
nativism (Fodor 1975, 1981). According to this view, nearly all concepts correspond-
ing to individual words in natural languages are innate, including the likes of LINGUINI,
CARBURETOR, BEATNIK, and QUARK. Notice that it isn’t just the sheer volume of innate con-

cepts that makes this view so outrageous—the thousands and thousands of concepts
120 Stephen Laurence and Eric Margolis

corresponding to actual and potential natural language words—but also the fact that
most of these concepts are clearly newcomers in human history, dependent on spe-
cific historical, cultural, and technological conditions for their appearance. Fodor once
tried to explain how it might be that so many innate concepts could be sitting around
unused for much of human history only to suddenly become active. His suggestion was
that they get “triggered” by innately specified environmental conditions whose occur-
rence might depend on highly contingent prior events. But this really is a singularly
implausible maneuver on his part, and Fodor’s view has found few if any advocates
even among nativists.3 Fodor’s radical concept nativism is not just an outlier position
within the empiricism-nativism debate, it is also an outlier position on the nativist side.4
What would a more reasonable concept nativist position look like? First, while
denying that all or virtually all concepts are innate, it would nonetheless embrace the
existence of a substantial number of innate concepts, including concepts that pick
out abstract categories. Second, it would embrace a variety of innate special-purpose
acquisition systems that are geared to particular conceptual domains. And third, it
would also embrace concepts acquired via relatively general-purpose innate acquisi-
tion systems—indeed, any tenable form of concept nativism will acknowledge that a
great deal of the conceptual system is acquired, at least in part, by such general-purpose
Which concepts and special-purpose acquisition systems should a nativist say are
innate? Ultimately, of course, this is an empirical question; there is no specific list
that a concept nativist must be committed to. In our view, though, likely contenders
include concepts and innate special-purpose systems associated with the representa-
tion of objects, physical causation, distance, movement, space, time, geometry, agency,
goals, perception, emotions, thought, biological kinds, life stages, disease, tools, preda-
tors, prey, food, danger, sex, kinship, group membership, dominance, status, norms,
morality, logic, and number. Note, however, that this isn’t to advocate the grossly
implausible claim that all concepts related to these domains are innate or acquired via
innate special-purpose systems. For example, any sensible form of concept nativism
would reject the idea that PASTA and BONBON are innate but might nonetheless accept
that there is an innate special-purpose acquisition system, or set of systems, that is
involved in the conceptualization of food. Likewise, concept nativists needn’t suppose
that the concepts for advanced mathematics are innate but may well maintain that
there are innate numerical representations of one kind or another.

3. Chomsky (1991) flirted with Fodor’s radical concept nativism, but even Fodor himself has
now rejected the view (Fodor 1998, 2008).
4. See, for example, Laurence and Margolis (2002) and Pinker (2007) for highly critical assess-
ments of Fodor’s radical concept nativism that are nonetheless quite congenial to nativism in
Concept Nativism and Neural Plasticity 121

Earlier we noted that it would be wrong to suppose that only empiricists hold that
concepts are learned. We can now see why. This is because nativism isn’t confined
to postulating innate concepts. Rather, a big part of concept nativism is its appeal to
innate special-purpose systems of acquisition, and these systems are often best under-
stood as learning systems. For instance, an innate special-purpose system for food
might support the learning of which items in the environment are to be eaten and
which are to be avoided, guiding food preferences and food-seeking behavior. Or an
innate special-purpose system for faces might support the learning of concepts of indi-
viduals. These hypothesized systems are very much in the business of learning about
the world, according to the nativist. They are just specialized for learning particular
information in a way that is highly constrained by the nature of the learning system.
To a large extent, then, the difference between nativism and empiricism isn’t whether
learning is central to human concept acquisition but rather their differing views of
how learning works. While empiricists take learning to be almost exclusively mediated
by innate general-purpose learning systems, nativists maintain that general-purpose
learning systems, though real and important, are not sufficient, and also postulate
numerous innate special-purpose learning systems, holding that they are central to
conceptual development, to categorization, and to other concept-involving higher-
level cognitive processes.
It is also worth emphasizing that nativists don’t deny the existence of relatively
domain-general concept acquisition, although nativists are likely to see even domain-
general acquisition as typically relying on special-purpose systems or constraints of
one kind or another. For example, socially mediated learning undoubtedly facilitates
the acquisition of a wide range of concepts, including concepts that are peculiar to a
learner’s culture. But on a nativist view, this learning is likely to be mediated by innate
systems for acquiring cultural norms (Sripada and Stich 2006), by innate systems that
are responsive to pedagogical cues (Csibra and Gergely 2011), and by innate biases
governing cultural transmission, including biases regarding who to imitate (Boyd and
Richerson 1985).
In the rest of the chapter, we take up the question of how the facts related to neural
structure and function bear on the evaluation of concept nativism, but before we do
that, we should briefly comment on the status of the argument we offer for concept
nativism—the argument from neural wiring—and how it fits into the larger case for the
nativist framework. This argument is intended as a nondemonstrative argument that
takes the form of an inference to the best explanation. And so we maintain that the
argument provides strong—though defeasible—support for concept nativism. We take
it that the status of concept nativism turns entirely on nondemonstrative arguments
of this kind. Empiricists sometimes write as though empiricism is the default position
and that nothing short of an incontestable proof for concept nativism should move us
away from this default (see, e.g., Prinz 2005, 2012). However, this outlook is misplaced.
122 Stephen Laurence and Eric Margolis

Given that empiricists and nativists disagree about what is clearly a factual question
about the structure of the mind, both sides are equally in need of evidence and argu-
ment to establish their view and must equally make the case that the balance of empiri-
cal considerations stands in their favor. Proofs have no more place here than they do
in other disputes about the workings of the mind. Any nondemonstrative argument
for concept nativism will also need to be considered in light of the other empirical
arguments that bear on the empiricism-nativism debate. Although we don’t have the
space in this chapter to examine other arguments, we believe that there are important
arguments for concept nativism drawing on evidence from developmental psychol-
ogy, animal psychology, evolutionary psychology, anthropology, and other fields, that
these arguments are mutually reinforcing, and that it is this total package that makes
concept nativism such an attractive view.5 Fortunately, though, the argument from
neural wiring offers considerable support for concept nativism even taken in isolation.
So while we don’t want to give the impression that concept nativism stands or falls
with this one argument, we do think the argument from neural wiring constitutes a
solid reason for favoring concept nativism all the same.

5.3 Plasticity as a Challenge to Nativism

As we noted above, plasticity is often seen as providing an argument against nativism,

not for it. We will begin, then, with the considerations that are generally thought to
show that neural plasticity poses a serious challenge to concept nativism.
Plasticity has become a catchall term for the many ways in which neural organiza-
tion and function change in response to an animal’s experience and action, as well as
to traumas to its brain and body. A standard example that is often used to illustrate
the general idea is the reorganization of the sensory map of the hand in the brain in
response to an injury or to changes in the way the hand is used. The normal arrange-
ment in the somatosensory cortex is for adjacent groups of cells to correspond to adja-
cent regions of the body—for example, neurons that respond to the index finger reside
close to neurons that respond to the middle finger. What happens if the nerve fibers
connecting a finger to the spinal cord are severed? One might expect the cortical area
that previously responded to that finger to atrophy, but this is not what happens.
Instead, this cortical area is taken over by the adjacent finger(s), changing the function
of that cortical area so that it responds to the adjacent finger(s). Likewise, even if one
of two adjacent fingers is simply used more or happens to receive a greater amount of
stimulation, some of the neurons that were originally responsive to the less active or
less stimulated finger become responsive to the more stimulated finger. As one major

5. For detailed discussion of a broad range of arguments for concept nativism and their interrela-
tions, see Laurence and Margolis (unpublished ms.).
Concept Nativism and Neural Plasticity 123

textbook summarizes the matter, “This functional plasticity suggests that the adult
cortex is a dynamic place where changes can still happen. Such phenomena demon-
strate a remarkable plasticity” (Gazzaniga, Irvy, and Mangun 2009, 102).
Neural plasticity doesn’t stop with potential changes to the boundaries of somato-
sensory representations. Even more interesting are instances in which cortical circuits
deprived of their usual sensory input come to be recruited by a differing sensory modal-
ity. An important example of this type of reorganization can be found in early-blind
Braille readers. When tested on tactile discrimination tasks, they show increased
activation in the visual cortex compared to sighted subjects, who show deactivation in
this area (Sadato et al. 1996).6 Studies using repetitive transcranial magnetic stimula-
tion (rTMS) further confirm the role of the visual cortex in the blind. This technique
employs a magnetic pulse to disrupt neural activity in a targeted region of the brain.
When rTMS is used to disrupt neural activity in the occipital (visual) cortex, blind
subjects have difficulty identifying Braille and embossed roman letters, whereas there
is no effect on sighted subjects engaged in comparable tactile discrimination tasks
(Cohen et al. 1997).
Perhaps the most celebrated instance of neural plasticity comes from a study with
ferrets in which ferrets’ brains were surgically rewired shortly after birth (Sharma,
Angelucci, and Sur 2000; von Melchner, Pallas, and Sur 2000). Retinal projections were
rerouted so that the neural signals that would normally go to the primary visual cortex
were fed to the primary auditory cortex via the auditory thalamus. When the ferrets
were tested as adults, not only did the auditory cortex in those with rewired brains
come to exhibit patterns of activity that are characteristic of the visual cortex (e.g., it
contained groups of cells that responded differentially to stimulus orientation), but the
ferrets were able to approach objects that could only be detected by sight. The take-
home message, according to one of the original research reports announcing these
results, is that “the pattern of early sensory activation can instruct the functional archi-
tecture of cortex to a significant extent” (Sharma, Angelucci, and Sur 2000, 846).
There is, of course, a question of how much of the brain exhibits this level of
plasticity. Is plasticity unique to sensory-perceptual systems? One reason to think it
isn’t is that there are cases in which children recover from focal damage to cortical
areas involved in language. Amazingly, there are even cases of children who come
to develop near-normal linguistic abilities after undergoing a hemispherectomy, in
which one cerebral hemisphere is disabled or entirely removed (Curtiss and Schaeffer
2005; Curtiss and de Bode 2003). One child (known as EB) who underwent a left

6. Early-blind subjects include those who are blind from birth as well as those who developed
blindness early in life. In the study by Sadato and colleagues, this includes children who were
blind before the age of seven, while in some of the other studies we cite below, the cutoff point
for being considered an early-blind subject is at significantly earlier ages.
124 Stephen Laurence and Eric Margolis

hemispherectomy at two and a half years old managed to recover much of his lan-
guage skills two years later.7 When tested at age fourteen, his language was found to
be normal in most respects, with all his linguistic functions now residing solely in his
right hemisphere (Danelli et al. 2013).
There may also be general theoretical reasons to suppose that the parts of the brain
that are involved in higher cognitive processes are plastic and hence can take on any
number of differing functions. Buller and Hardcastle (2000) argue that this is extremely
likely given that dedicated brain circuits per se don’t exist even for “our most basic [i.e.,
sensory] processes” (313). In general, they claim, “the dedication of a brain system to a
particular task domain is subject to change as the inputs to that brain system change”
(313). The reason there is the appearance of isolable and stable neural structures is that
“plastic human brains have encountered recurrent environmental demands through-
out history” (317). Elman et al. (1996) express a similar view. After reviewing the
evidence regarding brain plasticity in animals and humans, they conclude that there
may be some “primitive innate representations in the midbrain, … but the cortex
appears to be an organ of plasticity, a self-organizing and experience-sensitive network
of representations that emerge progressively across the course of development” (315).
It is worth noting that singling out the cortex isn’t uncommon among empiricists
who place a lot of weight on brain plasticity. Quartz (2003), for example, disagrees with
Elman and colleagues about subcortical structures, maintaining instead that innate
neural systems are the norm in most animals. But he agrees that a significant level
of plasticity is a peculiar feature of the human cortex, pointing, in particular, to the
example of how the occipital cortex functions differently in blind Braille readers than
in sighted control subjects (as in the work by Sadato et al. 1996 described above).
According to Quartz, “The sharp contrast between cortical and subcortical structures
suggests that the evolution of cortex may represent the evolution of a new acquisition
strategy” (Quartz 2003, 36).
In sum, there is a lot to be said for the view that the brain exhibits a great deal
of plasticity in terms of its functional structure. It also isn’t difficult to see why this
fact is often thought to favor empiricist approaches to conceptual development. Wide-
spread and significant instances of neural plasticity suggests an inherent openness to
the functions that any cortical area can take on. If this is right, then the brain’s concept
acquisition capacities needn’t be innately constrained toward any particular outcome.
Instead, cortical circuits might simply form as required to accommodate a learner’s
needs given whatever contingent sensory input has been received and the wiring that
has been previously established. Buller and Hardcastle (2000) capture this picture in
suggesting that neural plasticity goes hand in hand with the idea of a content-neutral

7. In right-handed individuals, like EB before his surgery, language is usually controlled by struc-
tures in the left hemisphere.
Concept Nativism and Neural Plasticity 125

capacity for addressing information-processing tasks. As they put it, “Our ancestors
may have encountered diverse problems, but the brain evolved a general solution to
those problems” (317; italics added).

5.4 The Argument from Neural Wiring

We have seen that the phenomenon of neural plasticity appears to constitute a major
objection to concept nativism. The more equipotential the brain, the less plausible it is
that there are innate special-purpose acquisition systems. And examples like the ones
reviewed in the last section do seem to point in the direction of a highly equipotential
brain with a functional organization that fits naturally with an empiricist approach
to cognitive and conceptual development. However, appearances are misleading. In
this section, we argue that not only do the facts pertaining to neural plasticity fail to
discredit concept nativism, but a careful look at the relevant neurological evidence
strongly favors the nativist approach. What we are calling the argument from neural
wiring is an argument for nativism that draws on a broad range of neurological evidence
showing that neural plasticity is highly constrained in ways that are best explained
within a nativist framework.

5.4.1 Plasticity Revisited

We begin with some observations about a few of the striking examples from the
last section. It is instructive to see that even in these instances, which are often thought
to make nativist views look hopeless, the changes that the brain undergoes aren’t as
flexible and open-ended as they first appear.
Consider the finding that language can be recovered after the loss of the cortical
areas that normally support linguistic processing, even with the loss of as much as
half of the cortex. We saw this with EB, who had his left cerebral hemisphere removed
when he was just two years old and yet grew up to have near-normal language abili-
ties. In his daily life, he doesn’t exhibit any noticeable signs of language impairment.
Clinical assessments have also found his language to be near normal (compared to
age-matched controls), with below average performance in only a few tests (e.g., in
reading, he had difficulty with homophones and words that have been borrowed from
other languages). These minor difficulties only highlight how strong his core linguis-
tic abilities are. EB is a particularly interesting case study among hemispherectomy
patients because the pathology he suffered didn’t involve epilepsy and was localized
in one hemisphere, leaving the other hemisphere intact and healthy. The pathologies
that lead to hemispherectomy aren’t usually so localized; generally there is damage to
both hemispheres.
At first glance, then, this may suggest that EB offers a clear-cut illustration of how
the plasticity of the brain stands in opposition to a nativist theory of development. EB’s
126 Stephen Laurence and Eric Margolis

right hemisphere took on cognitive functions that are usually located elsewhere in the
brain—a massive relocation of the neural circuitry for language (and for much else).
But functional magnetic imaging (fMRI) shows that the areas in his right hemisphere
supporting his linguistic abilities aren’t scattered in unpredictable ways, as one might
expect if his recovery were based on the powers of a truly equipotential brain. Rather,
EB’s language areas are homologs of the left hemisphere areas activated in linguistic
tasks in healthy control subjects.8 Danelli et al. (2013) performed an in-depth fMRI
analysis of language production versus comprehension, automatic versus controlled
language processes, and auditory versus visual processing, finding that “the overall
neurofunctional architecture of EB’s right hemispheric language system mirrors a
left-like linguistic neural blueprint” (225). This result points to a highly constrained
neural organization that illustrates an important general principle: the brain’s two
hemispheres incorporate a large measure of potential redundancy of function that can
be exploited at certain stages of development. This is a form of plasticity, to be sure, but
not a kind that favors empiricism.
What about the ferrets? Certainly this at least illustrates true equipotentiality?
After all, their rewired brains led to the auditory cortex developing features of the
visual cortex, and even to the ferrets being able to respond to visual stimuli relying on
processing that could only occur in their auditory cortex. Surprisingly, even this case
fails to provide a strong argument against nativist theories of conceptual and cognitive
For one thing, generalizing from one relatively small area of the auditory cortex
to the rest of the brain is a huge leap (Pinker 2002). Even if it turns out that this
portion of the auditory cortex and its downstream perceptual processing is highly
malleable in a way that reflects a given input modality, it doesn’t follow that the
processing for higher cognitive functions can be handled by arbitrary cortical areas.
Higher cognitive processes may well involve dedicated brain areas even if sensory pro-
cesses do not.

8. Contrary findings in clinical and neuroimaging studies have led to a controversy about
whether there are language-specific neural circuits—circuits that are specialized for and selec-
tively activated by language processing. On the one hand, clinical studies have found sharp
dissociations between linguistic and nonlinguistic deficits, suggesting the existence of language-
specific areas of the brain. But on the other hand, neuroimaging studies with neurotypical sub-
jects have failed to show processing regions exclusively devoted to language processing. The
problem with the neuroimaging work that fed this controversy, however, is that the selection of
the brain regions for examination didn’t take into account individual differences in neuroanat-
omy. Once the classic language areas are singled out in a way that is sensitive to these differences,
fMRI studies reveal the existence of neural regions selectively activated by linguistic processes to
the exclusion of other cognitive processes (such as mental arithmetic and general working
memory) in accordance with the clinical data (Fedorenko, Behr, and Kanwisher 2011).
Concept Nativism and Neural Plasticity 127

In the previous section, we encountered an argument by Buller and Hardcastle sug-

gesting the opposite conclusion. Their claim was that sensory processes are among
the most “basic” in the cortex, and consequently, that if there are no dedicated brain
areas for these processes, then it is even less likely that there would be dedicated
areas for higher cognitive processes. However, a lot is being packed into the idea of a
so-called basic psychological process in this argument. It may be true that, when the
brain receives external environmental stimulation, sensory processes are typically
activated earlier than higher cognitive processes. However, this isn’t enough to get Buller
and Hardcastle’s argument off the ground. What they need is for sensory processes to be
more fundamental than higher cognitive processes in ontogeny—for the development
of the neural systems for cognitive processes to depend on the prior development of, and
input from, the neural systems for sensory processes. But this claim would be question-
begging in the present context. Whether there is such a dependence is partly what
empiricists and nativists disagree about when the debate turns to the wiring of the brain.
Moreover, the point about the danger of an overgeneralization isn’t just about
the divide between sensation and cognition. It also applies to the internal structure
of perceptual systems as well. The effect of the rewiring experiments was that the
primary auditory cortex in a ferret with a rewired brain was able to process information
from the retina that normally would have been processed by the primary visual cortex,
and consequently came to take on some of the features that are characteristic of the pri-
mary visual cortex. The primary visual cortex is connected to a large number of distinct
brain regions that support further specific types of visual processing, including compu-
tations responsible for downstream representations of location, direction of motion,
speed, shape, and so on. If the redirected sensory input in the ferrets had led to the
development of all this downstream structure in the auditory cortex, that would amount
to exactly the sort equipotentiality of cortical areas that empiricists could use against
nativists. But, in fact, none of this downstream structure was reproduced. The overall
wiring of the ferrets’ auditory cortex was largely unchanged (Majewska and Sur 2006).
This situation—in which the primary auditory cortex in some sense allowed the
ferrets to “see” despite the fact that downstream auditory areas weren’t significantly
altered—leads to a puzzle. How is it that the primary auditory cortex can process and make
any use of visual information? Interestingly, Sur, the team leader in the ferret study,
has answered that “the animals with visual inputs induced into the auditory pathway
provide a different window on some of the same operations that should occur nor-
mally in auditory thalamus and cortex” (Sur 1988, 45; quoted in Pinker 2002, 96). In
other words, the processing that is supposed to be handled by these different sensory
areas is somewhat similar. Pinker expands on this idea by noting that there are general
high-level likenesses between the computations that might be expected to take place in
hearing and vision, as sound-makers with different pitches may be treated like objects
in different locations, and sharp changes in pitch may be treated like motions in space
128 Stephen Laurence and Eric Margolis

(Bregman and Pinker 1978). If this is right, then even though the rewiring experiments
show that the auditory cortex can be recruited for a certain amount of visual process-
ing, this is because the auditory cortex and the visual cortex overlap in the types of
computations they naturally support. Far from being a model case of the environment
instructing an equipotential cortex, Sur and colleagues’s rewiring experiments illustrate
the way in which cortical structure and function remain largely unchanged even in the
extreme case of input coming from a different sensory system.

5.4.2 Constrained Plasticity in Neural Structural Organization

So far, our response to the appeal to plasticity as an argument against concept nativ-
ism has been to show that some of the flagship examples of plasticity don’t really
amount to the sort of flexibility of neural organization that would tell against the
nativist approach to cognitive and conceptual development. We now turn to evidence
related to the wiring of the brain that directly supports the nativist perspective. The
general form of our argument is to point to aspects of neural, cognitive, and conceptual
development that exhibit constrained plasticity—development that is not open-endedly
plastic, but instead is highly constrained in ways that suggest important innate biases,
predispositions, and limits on neural structure and function. We begin, in this section,
by briefly presenting some evidence indicating that neural structural organization is
not primarily driven by environmental input or feedback configuring the brain.
In a landmark investigation, Verhage et al. (2000) examined neurological develop-
ment in a group of mutant (or knockout) mice whose brains (as a result of the genetic
mutation) were unable to release any neurotransmitters and thus were deprived of all
synaptic transmission.9 Accordingly, these mice would have had no experience-driven
neural development whatsoever. Verhage and colleagues compared these mice to con-
trol littermates and found that up until birth, their brains were remarkably similar.10
As these researchers explain:

Despite the general, complete, and permanent loss of synaptic transmission in the knockout
mice, their brains were assembled correctly (Fig. 3). Neuronal proliferation, migration, and dif-
ferentiation into specific brain areas were unaffected. At E12 [embryonic day 12], brains from
null mutant and control littermates were morphologically indistinguishable (Fig. 3, A and B). …
At birth, late-forming brain areas such as the neocortex appeared identical in null mutant and
control littermates, including a distinctive segregation of neurons into cortical layers (Fig. 3, C
and D). Furthermore, fiber pathways were targeted correctly in null mutants … (Fig. 3, G and H).
(Verhage et al. 2000, 866; figure references are to figure 3 in the original article, reproduced here
as figure 5.1, plate 9.)

9. Specifically, Verhage and colleagues suppressed the expression of the munc18-1 gene in the
10. After birth, of course, the mutant mice died. Without functioning synaptic communication,
the brain can’t support even the most basic life functions, such as breathing.
Control Null Control Null

Figure 5.1 (plate 9)

Neurological development in mutant mice that were genetically engineered to eliminate synaptic
transmission (null) and in normal control mice (control). (From Verhage et al. 2000. Used with
130 Stephen Laurence and Eric Margolis

This degree of similarity shows that many features of even the fine-grained structure
of the brain can develop without any sensory input or feedback. Experience-driven
neural activity may play more of a role in fine-tuning and maintaining this layout than
in establishing the overall organization itself (Marcus 2004). None of this is to say that
the brain isn’t plastic to some degree.11 But it underscores the fact that neural develop-
ment isn’t open-endedly plastic.

5.4.3 Constrained Plasticity in Functional Organization

Constrained plasticity isn’t confined to neural structural organization. It also extends
to the functional specificity associated with particular cortical areas. This is especially
apparent in instances of congenital and early sensory deficits, such as blindness,
in which the specific cognitive function of the affected cortical area is preserved.
Preserved function in these cases argues for constrained plasticity because it shows that
the information processing associated with the region isn’t dictated by the sensory
information it receives.
Viewed in a certain light, congenital sensory deficits are essentially rewiring experi-
ments. Cortical areas that usually process sensory information from an impaired
modality (e.g., subareas of the visual cortex in the congenitally blind) end up taking
input from another sensory modality, an organization (or reorganization) known as
cross-modal plasticity. An empiricist who supposes that the cortex has a high degree
of equipotentiality should predict that the resulting functions of these cortical areas
would differ from the normal case, reflecting the difference in the input. But recent
work with congenitally blind subjects shows that, contrary to this prediction, what
typically happens is that downstream components of the visual cortex and related
brain areas have the same functional specificity in the congenitally blind as in sighted
Consider, for example, the representation of the spatial location of objects in the
visual cortex. In one study, Renier et al. (2010) presented early-blind subjects with
auditory stimuli that varied in terms of sound type (different piano chords) and
spatial location.12 Using fMRI, their brain activity was measured during two behav-
ioral conditions, an identification condition (in which they had to determine whether
sequentially presented stimuli were of the same sound type) and a location condition
(in which they had to determine whether they had the same location). Renier and
colleagues found that the anterior part of the right middle occipital gyrus (MOG)—a

11. Notice that it shouldn’t be controversial that the brain is plastic in the minimal sense that
changes to the brain occur as people learn, think, and experience the world. This is simply a
requirement of any broadly materialist theory of the mind.
12. In this experiment, the early-blind subjects were either blind from birth or by the second
year of life, without ever having had normal vision or, at the time of testing, memories of visual
Concept Nativism and Neural Plasticity 131

part of the visual cortex associated with the representation of visual spatial location
in sighted subjects—was differentially active for the auditory spatial localization task
relative to the auditory identification task. The researchers also ran an analogous tactile
task with the same subjects. In this case, their fingertips were given different types of
stimulation (for the identification task), or there was stimulation to different fingers
(for the spatial location task). Once again, fMRI data revealed that the right anterior
MOG was differentially active for the spatial localization task relative to the identifica-
tion task.13 The upshot of this study is that, while the MOG is clearly plastic—in the
early blind it comes to subserve auditory and tactile spatial localization abilities that
it does not subserve in sighted individuals—the plasticity it exhibits is a form of con-
strained plasticity. The MOG continues to carry out the function of spatial localization
in the early blind, just with different types of sensory input.
Further fMRI studies have revealed the same pattern of constrained plasticity in
other neural regions with associated spatial functions. For example, Lingnau et al.
(2014) examined the activity in the posterior parietal cortex (PPC), which is involved
in the representation of space for purposes of guiding action. The PPC normally takes
its sensory input primarily from vision and exhibits a pronounced gradient—with pos-
terior subregions recruited more heavily for (visual) guidance of reaching and grasping,
and anterior subregions more for planning and execution of motor action. Using a
proprioceptively guided reaching task, however, Lingnau and colleagues were able to
show that the same pattern of functional differentiation occurs in congenitally blind
subjects. The researchers compared the brain activity of congenitally blind subjects
and blindfolded sighted subjects performing one of two actions on an object (touching
with fingertips vs. grasping with their whole hand) in a specified location (chosen from
five possible locations). In this case, the principle of constrained plasticity predicts that
sighted and blind subjects would have similar activation in the anterior portions of
the PPC (since all subjects would be equally engaged in the planning and execution
of motor action), but that there would be significantly greater activation in the
posterior portions of the PPC in the blind (since in the blind, the PPC would be accom-
modated to nonvisual sources of information regarding spatial location). This is exactly
what Lingnau and colleagues found, leading them to conclude that “neural plasticity
acts within a relatively rigid framework of predetermined functional specialization”
(Lingnau et al. 2014, 547).
One of the major features of the visual cortex is the functional division correspond-
ing to two broad networks of interrelated neural regions. The ventral visual stream
(the what pathway) represents object properties, and is involved in object recognition;

13. Blindfolded sighted control subjects doing these auditory and tactile tasks did not show the
same activation of the MOG. However, MOG activation in sighted subjects did occur in a compa-
rable visual task.
132 Stephen Laurence and Eric Margolis

the dorsal visual stream (the where pathway) represents object location and the spatial
relations between objects, and is involved in object-directed action. The results from
Renier et al. (2010) and Lingnau et al. (2014) that we have been reviewing indicate
that the dorsal visual stream continues to exist in early-blind and congenitally blind
subjects and that its component subregions engage in the same functional processing
for object location despite profound changes in sensory input (auditory or propriocep-
tive vs. visual).
Further studies of the dorsal visual stream fill out this picture by showing that it’s
not just the representation of spatial location that is preserved. For example, Wolbers,
Zahorik, and Giudice (2011) examined activity in the dorsal occipitotemporal cortex
in congenitally blind adults, focusing on a region of interest encompassing the hMT+
complex, which normally represents the direction of visual motion. To determine
whether this region retains the same function when deprived of its usual (visual) input,
an fMRI scan was taken while congenitally blind subjects heard leftward and rightward
broadband noise signals, as well as static control stimuli. Wolbers and colleagues found
that the region of interest was specifically involved in motion detection in congenitally
blind subjects even though the sensory input in this case was auditory, not visual.14
Once again, we have an impressive instance of plasticity (the fact that the dorsal visual
pathway is co-opted for auditory processing), but the plasticity is constrained, preserv-
ing the normal functional specificity of a dorsal pathway subregion.
Constrained plasticity has been found in the ventral visual pathway as well. For
example, Striem-Amit et al. (2012) trained congenitally blind subjects and sighted
controls to use a sensory substitution device, which transforms visual information from
a head-mounted camera into auditory information so that soundscapes can be used to
detect visual stimuli. Blind and (blindfolded) sighted subjects had up to an hour and
a half of training on the device before being tested on simple geometric shapes in
different locations. Upon hearing the prompt “shape,” they had to judge whether the
stimulus was circular or angular; upon hearing “location,” they had to judge whether it
was on the left or the right side. fMRI data showed that both the blind and the sighted

14. Currently, far more studies are directed to visual impairments than to other sensory impair-
ments, but there is evidence that the same overall pattern holds when the impairment is to
cortical areas that usually draw on nonvisual input. For example, work with deaf cats has shown
that just as the visual cortex can process auditory or tactile stimuli in the congenitally blind, the
auditory cortex can process visual stimuli in the congenitally deaf. Lomber, Meredith, and Kral
(2010) examined congenitally deaf cats, which possess enhanced visual location abilities. By tem-
porarily deactivating differing portions of the auditory cortex (using surgically implanted cooling
loops), these researchers were able to determine that the cortical area responsible for this
enhancement is the posterior auditory field, a region involved in the localization of acoustic
stimuli in hearing cats.
Concept Nativism and Neural Plasticity 133

groups exhibited differential activation for location versus shape, with activation
for shape in the inferior temporal cortex (in the ventral visual stream), and activa-
tion for location in the precuneus and middle temporal sulcus/gyrus (in the dorsal
visual stream). What’s more, in blind subjects, there was increased activation for shape
information in the ventral visual stream (in ventral Brodmann area 19), suggesting
that prior use of the ventral visual stream for auditory stimuli enhanced this func-
tion. Thus it would appear that the large-scale functional architecture of the visual
cortex—the division of labor between the dorsal and ventral streams—develops in
much the same way, and with the same functions being performed in various subre-
gions of these streams, with or without visual experience.
These and related studies showing preserved functional specificity in the visual
cortex support what is known as the metamodal hypothesis regarding the brain’s func-
tional organization (Pascual-Leone and Hamilton 2001). According to this hypothesis,
much of the brain is composed of distinct computational systems whose functions are
established independently of their sensory input. These systems are capable of pro-
cessing sensory information from differing modalities but settle on a given modality
when the input it provides is the best fit for the computations carried out—thus giv-
ing the appearance of modal specificity. On this view, it is a misnomer to speak of the
“visual cortex,” the “auditory cortex,” and so forth. Rather, each of these broad areas
is composed of neural systems that engage in computations that create a preference
for a given modality, but the computations performed aren’t inherently about visual
or auditory content, so when the preferred input is unavailable, the brain switches to
the next best fit. As our discussion above suggests, there is now a considerable amount
of evidence in support of the metamodal hypothesis. For our purposes, though, what
matters is the implication for nativism. Notice that to the extent that the brain is orga-
nized in this way, we have grounds to suppose that the functional specificity associated
with particular regions of the brain is innate. The reason the hMT+ computes direction
of motion, for example, can’t be because this is required by its visual input; it performs
the same function in the complete absence of visual input in the congenitally blind.
Rather, the most plausible explanation of its functional specificity is that this brain
region is innately organized for computing direction of motion, and this results in it
selecting visual input when visual input is available because visual input is optimal for
the computations it performs.
Some of the examples of innate neural systems we have been reviewing may not
seem especially conceptual. However, whether they involve conceptual representations
or not, they do at least contribute to the formation and processing of representations
that undoubtedly are—for example, concepts of movement, location, and spatial
relations—and thereby count as part of the arrangement of special-purpose systems
that explains concept acquisition for nativists. In any case, other work speaks more
directly to the neural basis of conceptual-level representations.
134 Stephen Laurence and Eric Margolis

For example, Mahon et al. (2009) examined the ventral visual stream’s representa-
tion of living versus nonliving kinds. It is well known that the ventral visual stream
exhibits neural specialization for these differing categories, with the representation of
artifacts (e.g., tools and nonmanipulable objects) in medial regions and the represen-
tation of living animate things (e.g., animals and faces) in lateral regions. A common
assumption in neuropsychology is that this medial-to-lateral organization stems from
the differing visual features associated with these categories, which are claimed to
lead to differing types of visual experiences with exemplars from these categories
(e.g., Rogers et al. 2005). One way to test this supposition is to compare sighted and
congenitally blind subjects using a common task that would be expected to generate
ventral visual stream activation in sighted subjects. This is exactly what Mahon and
colleagues did, asking sighted and congenitally blind subjects to make size judgments
upon hearing words for artifacts and animals. In both blind and sighted subjects, the
same medial-to-lateral organization was found. Now, if representations of living and
nonliving kinds were organized as they are in the ventral visual stream of sighted
subjects because of a response to some measure of visual similarity, it would be deeply
surprising to find the same fine-grained functional differentiation along the medial-to-
lateral axis among the congenitally blind. As these researchers note, the data suggest
instead “that the organization of the ventral stream innately anticipates the different
types of computations that must be carried out over objects from different conceptual
domains” (Mahon et al. 2009, 403).
It is certainly noteworthy that the organization of high-level representations in the
visual cortex is retained notwithstanding the complete lack of visual input. But just
as important are instances of higher cognitive amodal neural systems retaining their
functional specificity despite a lack of visual input. After all, these systems still depend
on sensory information, so the information they draw on will differ in dramatic ways
when visual information is not available. Consider, for example, the effect of blind-
ness on the development of ordinary mentalizing abilities, which include the ability to
attribute mental states to others and to oneself, and the ability to reason about mental
states and their role in behavior. Blind individuals lack access to many of the perceptual
cues that are typically associated with these abilities, such as others’ facial expressions,
direction of gaze, and body posture. Blind individuals also can’t rely on first-person
experience to understand other people’s visual perception of events. Despite these radi-
cal differences, however, the location of the neural substrates for mentalizing in early-
blind individuals (including congenitally blind individuals) is the same as for sighted
individuals (Bedny, Pascual-Leone, and Saxe 2009). Notice how unexpected this is on
the assumption that mentalizing is acquired largely on the basis of general-purpose
processes that are especially sensitive to perceptual cues. Why would the same cortical
areas end up with the same peculiar functions given such grossly different access to
the evidence for mental activity? In contrast, this constancy in function is naturally
explained on the hypothesis that these cortical areas and the functions they realize are
Concept Nativism and Neural Plasticity 135

determined independently of perceptual input, reflecting a psychological capacity that

is, to a significant extent, innate.15
In sum, there is considerable evidence for preserved functional specificity of
neural areas even when an early sensory deficit promotes cross-modal plasticity. If
the cortex were an equipotential network, the brain ought to undergo an immense
functional reorganization in cases of congenital and early blindness. But that’s not
what happens at all. We’ve seen the preservation of the fine-grained functional
structure in what is usually thought of as a visual-motor area, in areas that repre-
sent spatial properties (such as location and direction of motion), in the large-scale
functional differentiation between the ventral and dorsal visual streams, in the
functional specificity within the ventral stream (which includes different areas for
representing artifacts and living kinds), and in amodal neural centers (e.g., areas
associated with mentalizing) that rely heavily on sensory input. Taken together
these and related studies constitute a diverse and compelling body of evidence for
constrained plasticity.16

15. Bedny, Pascual-Leone, and Saxe (2009) offer a qualification to these conclusions, citing work
that suggests that an understanding of false belief develops at a later age in blind individuals than
in sighted individuals, perhaps as late as eight years old (e.g., Peterson, Peterson, and Webb
2000). However, this work is based on what has come to be called elicited-response false-belief tasks
(where children are asked direct questions regarding the mental states of others), rather than
spontaneous-response false-belief tasks (where children display their knowledge of others’ mental
states in their spontaneous responses to false beliefs in others) (Baillargeon, Scott, and He 2010).
Since sighted children have been shown to pass spontaneous-response tasks at much younger
ages than elicited-response tasks (Onishi and Baillargeon 2005; Surian, Caldi, and Sperber 2007;
Luo 2011; Kovács, Téglás, and Endress 2010), it would be very interesting and revealing to deter-
mine if blind infants can pass nonvisual, nonverbal spontaneous-response false-belief tasks at a
comparable age to sighted infants. As far as we know, though, all spontaneous-response false-
belief tasks that have been run on infants to date have been visually-based tasks, so new tasks
would need to be designed to test this possibility.
16. Another case of preserved functional specificity in the face of sensory deficits can be found in
an area of the visual cortex that is involved in word and letter recognition in reading. The neural
region subserving this ability is in fact remarkably stable across individuals and languages. Meta-
analyses of numerous studies show that “the same region of the left lateral occipitotemporal
sulcus always is activated, to within a few millimeters, whenever literate humans read” (Dehaene
and Cohen 2011, 256). This case is particularly interesting in light of the fact that reading is such
a recent cultural invention; there couldn’t be a biological adaptation for reading per se. Rather, the
most plausible explanation of this functional specificity—a broadly nativist explanation—is that
the abstract categorization of words and letters is always subserved by this neural region because
it has an innate computational structure and functional connectivity that makes it uniquely well
suited to playing this role, and that the cultural practice of reading has been altered in history to
better accommodate the peculiarities of the computational processes that it did evolve for
(Dehaene 2009). Similar considerations address the neural plasticity in blind Braille readers.
136 Stephen Laurence and Eric Margolis

5.4.4 Constrained Plasticity in Cognitive and Conceptual Impairments

We turn now to a second type of case that argues in favor of the innate functional
specificity of particular brain regions: cognitive and conceptual impairments due to
focal brain damage and genetic anomalies. Such impairments argue for constrained
plasticity because they show that the brain is unable to compensate for certain dif-
ficulties despite ample opportunity for neural reorganization and exposure to relevant
features of the environment.
Neurological disorders typically don’t affect a single functional system in isolation
but rather involve a variety of co-occurring deficits. For example, a stroke may result
in damage to functionally distinct yet physiologically neighboring brain areas that
are equally dependent on the impeded blood flow. Nonetheless, cognitive deficits
are sometimes quite specific. For example, prosopagnosia, a deficit in the ability to recog-
nize faces, may be accompanied by other forms of agnosia but can also occur as a selective
deficit in which the impairment is peculiar to faces. In this case, individuals may be unable
to recognize altered versions of faces yet have no difficulty in recognizing comparably
altered complex objects that aren’t faces (Busigny, Graf, et al. 2010; Busigny, Joubert, et
al. 2010; see also Rezlescu et al. 2014). This sort of specificity regarding a representational
deficit can persist despite many years of exposure to relevant stimuli and a strong vested
interest in the subject domain. Individuals who suffer from prosopagnosia, for example,
will often have as much difficulty recognizing faces of familiar and emotionally sig-
nificant people in their lives (parents, children, partners) as they have with strangers.
In addition, category-specific deficits in semantic memory (memory related to general
knowledge) aren’t tied to any particular type of task or a given modality (Capitani
et al. 2003). A selective deficit for the category of living kinds, for instance, may show
up equally in an inability to recognize animals in a picture-naming task and in purely
verbal queries about the features of different animals.
The specificity of a category-specific deficit might be explained in a number of dif-
ferent ways. We consider two broad classes of explanation regarding deficits in seman-
tic memory, focusing on the well-studied example of category-specific deficits in the
representation of living kinds—for example, patients with significant impairments for
animals (elephant, duck, etc.) in contrast with artifacts (pen, key, etc.). The standard
empiricist explanation of such cases holds that semantic memory isn’t organized in
terms of a categorical distinction between the living and the nonliving (or animal vs.
artifact), but instead is organized in terms of the properties that exemplars of particular
categories possess. Different types of properties are taken to figure more prominently in
the representation of categories of living versus nonliving kinds. For instance, on one
influential account, visual properties are taken to be more prominent for living kinds,
and functional properties for nonliving kinds. If this account were correct, then focal
damage to the neural substrate for the representation of visual properties would dis-
proportionately affect living kinds, while damage to the representation of functional
Concept Nativism and Neural Plasticity 137

properties would disproportionately affect nonliving kinds (Warrington and McCarthy

1983; Farah and McClelland 1991). The nativist approach, in contrast, maintains that
semantic memory is organized in terms of a categorical distinction between living and
nonliving kinds, and that in general there are innately dedicated neural circuits related
to a number of fundamental category types with particular evolutionary significance,
such as animals, tools, faces, and food (Caramazza and Shelton 1998; Mahon and
Caramazza 2009).
As our interest is in the empiricism-nativism debate, a particularly important type
of case to consider in evaluating these two different types of explanations is one in
which a category-specific deficit results from neural damage or from a genetic disorder
that affects early development. Farah and Rabinowitz (2003) documented the case of
Adam, who sustained brain damage when he was just one day old. At age sixteen,
Adam was tested for his knowledge of living and nonliving kinds, and a significant
difference between the two was found. Adam had a severe impairment for knowledge
regarding living kinds (responding to testing at chance levels), yet his performance
was normal or near normal regarding nonliving kinds. His difficulty with living kinds
was also comprehensive in that it affected visual and nonvisual properties alike,
while his knowledge of nonliving kinds (both visual and nonvisual) was spared. Con-
sequently, Adam’s psychological profile conflicts with the empiricist explanation of
category-specific deficits in terms of selective damage to the representation of a given
type of property (in this case, to visual properties).
What’s more, Adam’s case speaks directly to the limitations on neural plasticity in
cognitive development. Even though the neural damage occurred very early in devel-
opment, and Adam had years of experience in infancy and childhood in which other
aspects of his psychological development proceeded normally, his brain wasn’t able to
compensate for the damage it had sustained. As Farah and Rabinowitz put it:

... phrased in terms of Adam’s surviving brain tissue, despite its adequacy for acquiring semantic
memory about nonliving things, it could not take over the function of semantic memory for liv-
ing things. This implies that prior to any experience with living and nonliving things, we are des-
tined to represent our knowledge of living and nonliving things with distinct neural substrates.
This in turn implies that the distinction between living and nonliving things, and the anatomical
localization of knowledge of living things, are specified in the human genome. (408)

In a related study, Farah et al. (2000) examined a different specific representational

deficit in the same subject, namely, Adam’s difficulty with faces. At the age of sixteen,
Adam had the classic profile of prosopagnosia—lesions in the occipitotemporal cor-
tex (bilaterally), with a severe impairment in the ability to recognize faces relative to
good, though not perfect, object-recognition abilities.17 As with the living/nonliving

17. “In everyday life he is unable to recognise faces, whereas his object recognition ability is fully
adequate for activities of daily living” (Farah et al. 2000, 122).
138 Stephen Laurence and Eric Margolis

distinction, this uneven cognitive profile raises the question of why other neural tissue
was unable to compensate for the damaged neural tissue—a striking lack of plasticity—
especially given the obvious importance of face recognition in daily life.
Now, there are a number of possible explanations for why the representation of faces
might be impaired, just as there are different possible explanations for the selective
impairment to the representation of living (or nonliving) kinds. Duchaine et al. (2006)
addressed this issue by examining another patient, Edward, who suffered from devel-
opmental prosopagnosia. In this study, the researchers took advantage of the oppor-
tunity to test on a single subject all of the alternatives to the nativist domain-specific
explanation that have appeared in the face-perception literature. Among the empiricist
explanations Duchaine and colleagues looked into were the possibilities that Edward
suffered from a general difficulty regarding the representation of individuals within a
category, a general difficulty with holistic processing, a general difficulty with config-
ural processing (i.e., representing the spacing between features), and a general difficulty
in acquiring expertise for object categories. For example, the configural-processing
explanation was evaluated by having Edward make same-different judgments for
photographs of faces and houses that had been digitally altered. The distance between
the eyes or windows was changed, or these features themselves were replaced with
similar features in the same relative spacing. In this case, Edward’s performance was
normal for detecting changes to houses, but three standard deviations below the mean
for detecting commensurate changes to faces. Likewise, the expertise hypothesis was
evaluated using corresponding face- and body-matching tests, in which the goal was
to identify which of two rotated faces or headless bodies matched a target. Here, too,
Edward had great difficulty with faces, but his performance with bodies was normal—
in fact, he scored in the high end of the normal range for body recognition. These and
the results from Duchaine and colleagues’ other tests indicate that Edward’s difficulty
was genuinely face specific, and consequently that there are face-specific developmen-
tal mechanisms that may be selectively impaired.
Edward’s impairment (unlike Adam’s) was most likely the result of a genetic anom-
aly.18 Though not all genetic disorders that result in representational deficits are as
focused as prosopagnosia—most result in uneven but predictable profiles of spared
conceptual abilities and impairments—they can still provide an excellent source of

18. Face-recognition ability is highly heritable. Polk et al. (2007) compared the patterns of neural
activity to faces and other stimuli in monozygotic and dizygotic siblings and found that the
activity was significantly more similar for monozygotic siblings for faces and places, but not for
pseudowords or chairs. Likewise, Wilmer et al. (2010) tested monozygotic and dizygotic twins
on a face-memory task and several control tasks for nonface memory. They found that “genetic
differences can account for most of the stable variation in face recognition ability in healthy
adults” and that this is specific to faces (Wilmer et al. 2010, 5239).
Concept Nativism and Neural Plasticity 139

evidence regarding the limits on the brain’s plasticity. For example, individuals with
Williams syndrome, a rare genetic disorder (Schubert 2009), show severe deficits in
certain types of reorientation tasks that rely on geometric representation19 but rela-
tively spared face-recognition abilities (Lakusta, Dessalegn, and Landau 2010; Bellugi
et al. 2000), and they have intact biological motion representation despite other types
of motion representation deficits (Jordan et al. 2002; Reiss et al. 2005).
One particularly well-studied and illuminating case is the impairment to men-
talizing abilities found in individuals with autism spectrum disorder (ASD).20 In a
classic early investigation, Baron-Cohen, Leslie, and Frith (1985) examined three groups
of children on a false-belief task—clinically normal preschool children, children with
Down syndrome, and high-functioning children with ASD. In their false-belief task
(known as the Sally-Anne task), subjects witness a protagonist (Sally) place a marble in
her basket, which is then moved (by Anne) to a box while Sally is away from the scene.
When Sally returns, subjects are asked where she will look for her marble. Baron-Cohen,
Leslie, and Frith found that clinically normal children and Down syndrome children
both answered correctly, saying that she will look in the basket (where Sally should
falsely think that it is), while children with ASD overwhelmingly gave the incorrect
response, saying that she will look in the box (where it actually is). Subsequent work
by Leslie and Thaiss (1992) showed that this failure is specific to the understanding
of belief and is not part of a general difficulty with understanding representation. Leslie
and Thaiss compared clinically normal preschool children with high-functioning
children with ASD—this time using both false-belief tasks and structurally similar
tasks with photographs and maps. (In a false-photograph task, for example, a Polaroid
photo is taken of an object in one location, and the object is moved before the
photo is developed. Then the question asked is where the object will be in the
photograph.) Leslie and Thaiss found that children with ASD who failed false-belief
tasks were able to pass false-photograph or false-map tasks. By contrast, children who
didn’t have ASD found the false-photograph and false-map tasks more difficult than
the false-belief task.
More recent work has found that children with ASD not only have difficulties with
elicited-response false-belief tasks (in which they are explicitly asked to respond to
a false-belief scenario), but are also unable to anticipate an actor’s actions when
presented with evidence of the actor’s false belief in a spontaneous-response task

19. In a typical reorientation task, subjects in a rectangular room are shown the hiding place for
an object and are gently spun around until they become disoriented. They are then asked to
locate the object, which requires using the geometry of the room to become reoriented.
20. Heritability studies indicate a strong genetic component in ASD, but it appears that there are
numerous different genetic anomalies that give rise to the characteristic impairments in ASD (see,
e.g., Huguet, Ey, and Bourgeron 2013).
140 Stephen Laurence and Eric Margolis

(Senju et al. 2010). This is not due to a general inability to understand action, as they
correctly predict an agent’s actions when the agent doesn’t have a false belief and are
able to correctly attribute goals to an agent even when the agent fails to achieve his or
her goal (Carpenter, Pennington, and Rogers 2001). Further, this sort of impairment per-
sists into adulthood. Senju et al. (2009) found that adults with ASD who can correctly
answer explicit questions about what an agent with false beliefs will do nonetheless
fail to spontaneously anticipate that an agent will act the same way in a live situation.
This suggests that they are solving elicited-response tasks using consciously formu-
lated rules that substitute for an intuitive understanding of the source of action. Other
work suggests a similar conclusion. Ordinarily people modulate their behavior when
they are observed because of the potential effect on their social reputation (e.g., giving
more money to a charity in the presence of others than when alone). In contrast, high-
functioning adults with ASD don’t modulate their behavior in this way (Izuma et al.
2011). Likewise, ordinarily people take into account the absence of negative intentions
when formulating a moral judgment pertaining to someone who accidentally causes
a negative outcome. Here, too, high-functioning adults with ASD behave differently,
treating cases with and without negative intentions equally (Moran et al. 2011).21 Thus,
a convergence of evidence suggests that ASD is associated with a selective represen-
tational impairment, one that affects the formation and use of certain mental state
concepts but not other concepts of comparable difficulty.

5.4.5 Summary and the Future

We began this section by reexamining the argument against nativist views of concepts
that appeals to considerations having to do with neural plasticity and showed that this
argument isn’t so compelling after all. We then presented a range of different types of
evidence for the opposing nativist hypothesis of constrained plasticity:

1. Evidence from preserved neural structural organization in the absence of relevant environ-
mental input This was seen in the genetically altered mice, whose brains developed
the same structural organization as their normal littermates even though the genetic
manipulations disrupted all synaptic transmission, eliminating any possibility for
sensory information or feedback to affect development.
2. Evidence from preserved neural functional specificity despite early sensory deprivation This
was seen, for example, in the preservation of the large-scale functional organization
of the visual system in the congenitally blind (the division of labor inherent to the
ventral and dorsal visual streams), and in cases where the functional and representational

21. The right temporo-parietal junction (rTPJ), which is known to be a critical mentalizing brain
area (Koster-Hale and Saxe 2013), is particularly involved in modulating moral judgments accord-
ing to whether a harm is accidental or intentional (Buckholtz et al. 2008; Young and Saxe 2009).
Interestingly, the normal spatially distinct responses within the rTPJ for accidental versus inten-
tional harms is absent in adults with ASD (Koster-Hale et al. 2013).
Concept Nativism and Neural Plasticity 141

specificity of particular neural subregions remains the same even though the input is
no longer visual (e.g., the assignment of the same neural region to the abstract repre-
sentation of living kinds).
3. Evidence from conceptual and cognitive deficits resulting from early focal neurological
trauma In this case, we saw that the brain’s plasticity is unable to compensate for such
deficits despite ample opportunity for neural reorganization and exposure to relevant
features of the environment (e.g., Adam’s inability to represent faces or living kinds in
4. Evidence from conceptual and cognitive deficits resulting from genetic anomalies The
brain’s plasticity is also unable to compensate for deficits resulting from developmental
genetic anomalies (e.g., spatial deficits associated with Williams syndrome and mental-
izing deficits associated with autism), despite ample opportunity for neural reorganiza-
tion and exposure to relevant features of the environment.

We conclude that the neural plasticity of the brain takes the form of constrained
plasticity—development that is not open-endedly plastic, but instead is highly con-
strained in ways that suggest important innate biases, predispositions, and limits on
structure and function.22
We want to reiterate, however, that the case for concept nativism doesn’t stand
entirely with the argument from neural wiring. Although what has been discovered
about the details of neural structure and function strongly supports the nativist view-
point, equally important are the findings in such fields as animal psychology, devel-
opmental psychology, evolutionary psychology, linguistics, and cross-cultural studies.
Moreover, the connection between work in these (and other) areas of cognitive science
and the argument from neural wiring is a relatively unexplored and potentially rich
source of insight about the origins of human concepts. For this reason, we suggest
that an exciting and important direction for future work on concepts is to revisit the
argument from neural wiring with the aim of looking for a far more extensive body of
innate systems of representation, guided by nativist hypotheses that are independently
motivated in these other areas of cognitive science.

5.5 Conclusion

Concept nativism holds that a significant number of concepts are either innate
or acquired via innate special-purpose acquisition systems. Concept nativism isn’t
opposed to learning; rather, it offers a distinctive perspective on learning, one that is

22. This list of different types of evidence for the hypothesis of constrained plasticity is not
intended to be exhaustive. Though space considerations prohibit an exploration of additional
types of evidence, two that are worth mentioning are (1) evidence from twin studies (see, e.g.,
footnote 18 above) and (2) evidence for parallel neurological structures subserving the same func-
tions across species (see, e.g., Kriegeskorte et al. 2008).
142 Stephen Laurence and Eric Margolis

grounded in the idea that much learning takes place only because the mind is innately
structured to extract specific types of information from the world and to process this
information in particular ways. Concept nativism tells us that we shouldn’t assume
there is a single general-purpose acquisition mechanism that accounts for the origin
of our concepts for animals, artifacts, mental states, individual people, and so on.
These and other content domains may well depend on distinct special-purpose innate
acquisition systems that are geared toward specific types of content.
The case for concept nativism takes the form of an inference to the best explana-
tion that draws on evidence from a range of disciplines. What we have shown here is
that one of the major objections to nativism—its alleged neurological implausibility in
the face of neural plasticity—is unfounded. Neural plasticity isn’t as flexible and open-
ended as nativism’s critics have supposed. On the contrary, the plasticity of the brain
is highly constrained in a way that argues for concept nativism. The brain is not com-
prised of an equipotential network that is sculpted into differentiated functional units
through sensory experience. Rather, the brain is innately differentiated into a complex
arrangement of distinct neural systems specialized for processing specific types of infor-
mation, exactly as concept nativism predicts.


We would like to thank Brad Mahon and Gerado Viera for their comments on an earlier
draft. Thanks also to Canada’s Social Sciences and Humanities Research Council and
the UK’s Arts and Humanities Research Council for supporting this research.


Baillargeon, R., R. M. Scott, and Z. He. 2010. False-belief understanding in infants. Trends in
Cognitive Sciences 14 (3): 110–118.

Baron-Cohen, S., A. M. Leslie, and U. Frith. 1985. Does the autistic child have a “theory of mind”?
Cognition 21 (1): 37–46.

Bedny, M., A. Pascual-Leone, and R. R. Saxe. 2009. Growing up blind does not change the
neural bases of theory of mind. Proceedings of the National Academy of Sciences of the United States of
America 106 (27): 11312–11317.

Bellugi, U., L. Lichtenberger, W. Jones, Z. Lai, and M. St. George. 2000. I. The neurocognitive
profile of Williams syndrome: A complex pattern of strengths and weaknesses. Journal of Cognitive
Neuroscience 12 (supplement 1): 7–29.

Boyd, R., and P. Richerson. 1985. Culture and the Evolutionary Process. Chicago: University of
Chicago Press.
Concept Nativism and Neural Plasticity 143

Bregman, A. S., and S. Pinker. 1978. Auditory streaming and the building of timbre. Canadian
Journal of Psychology 32 (1): 19–31.

Buckholtz, J. W., C. L. Asplund, P. E. Dux, D. H. Zald, J. C. Gore, O. D. Jones, and R. Marois. 2008.
The neural correlates of third-party punishment. Neuron 60 (5): 930–940.

Buller, D. J., and V. Hardcastle. 2000. Evolutionary psychology, meet developmental neurobiol-
ogy: Against promiscuous modularity. Brain and Mind 1 (3): 307–325.

Busigny, T., M. Graf, E. N. Mayer, and B. Rossion. 2010. Acquired prosopagnosia as a face-specific
disorder: Ruling out the general visual similarity account. Neuropsychologia 48 (7): 2051–2067.

Busigny, T., S. Joubert, O. Felician, M. Ceccaldi, and B. Rossion. 2010. Holistic perception of the
individual face is specific and necessary: Evidence from an extensive case study of acquired
prosopagnosia. Neuropsychologia 48 (14): 4057–4092.

Capitani, E., M. Laiacona, B. Z. Mahon, and A. Caramazza. 2003. What are the facts of
category-specific deficits? A critical review of the clinical evidence. Cognitive Neuropsychology

Caramazza, A., and J. R. Shelton. 1998. Domain-specific knowledge systems in the brain: The
animate-inanimate distinction. Journal of Cognitive Neuroscience 10:1–34.

Carpenter, M., B. F. Pennington, and S. J. Rogers. 2001. Understanding of others’ intentions in

children with autism. Journal of Autism and Developmental Disorders 31 (6): 589–599.

Chomsky, N. 1991. Linguistics and cognitive science: Problems and mysteries. In The Chomskyan
Turn, ed. A. Kasher, 26–53. Oxford: Blackwell.

Cohen, L. G., P. Celnik, A. Pascual-Leone, B. Corwell, L. Faiz, J. Dambrosia, M. Honda, et al. 1997.
Functional relevance of cross-modal plasticity in blind humans. Nature 389 (6647): 180–183.

Csibra, G., and G. Gergely. 2011. Natural pedagogy as evolutionary adaptation. Philosophical
Transactions of the Royal Society of London: Series B, Biological Sciences 366:1149–1157.

Curtiss, S., and S. de Bode. 2003. How normal is grammatical development in the right hemi-
sphere following hemispherectomy? The root infinitive stage and beyond. Brain and Language 86
(2): 193–206.

Curtiss, S., and J. Schaeffer. 2005. Syntactic development in children with hemispherectomy: The
I-, D-, and C-systems. Brain and Language 94 (2): 147–166.

Danelli, L., G. Cossu, M. Berlingeri, G. Bottini, M. Sberna, and E. Paulesu. 2013. Is a lone right
hemisphere enough? Neurolinguistic architecture in a case with a very early left hemispherec-
tomy. Neurocase 19 (3): 209–231.

Dehaene, S. 2009. Reading in the Brain. New York: Viking.

Dehaene, S., and L. Cohen. 2011. The unique role of the visual word form area in reading. Trends
in Cognitive Sciences 15 (6): 254–262.
144 Stephen Laurence and Eric Margolis

Duchaine, B. C., G. Yovel, E. J. Butterworth, and K. Nakayama. 2006. Prosopagnosia as an impair-

ment to face-specific mechanisms: Elimination of the alternative hypotheses in a developmental
case. Cognitive Neuropsychology 23 (5): 714–747.

Elman, J. L., E. A. Bates, M. H. Johnson, A. Karmiloff-Smith, D. Parisi, and K. Plunkett.

1996. Rethinking Innateness: A Connectionist Perspective on Development. Cambridge, MA: MIT

Farah, M. J., and J. L. McClelland. 1991. A computational model of semantic memory impair-
ment: Modality specificity and emergent category specificity. Journal of Experimental Psychology:
General 120 (4): 339–357.

Farah, M. J., and C. Rabinowitz. 2003. Genetic and environmental influences on the organisation
of semantic memory in the brain: Is “living things” an innate category? Cognitive Neuropsychology
20 (3–6): 401–408.

Farah, M. J., C. Rabinowitz, G. E. Quinn, and G. T. Liu. 2000. Early commitment of neural
substrates for face recognition. Cognitive Neuropsychology 17 (1–3): 117–123.

Fedorenko, E., M. K. Behr, and N. Kanwisher. 2011. Functional specificity for high-level linguistic
processing in the human brain. Proceedings of the National Academy of Sciences of the United States of
America 108 (39): 16428–16433.

Fodor, J. A. 1975. The Language of Thought. Tomas Y. Crowell.

Fodor, J. A. 1981. The present status of the innateness controversy. In Representations: Philosophi-
cal Essays on the Foundations of Cognitive Science. Cambridge, MA: MIT Press.

Fodor, J. 1998. Concepts: Where Cognitive Science Went Wrong. Oxford: Oxford University Press.

Fodor, J. A. 2008. LOT2: The Language of Thought Revisited. Oxford: Oxford University Press.

Gazzaniga, M. S., R. B. Irvy, and G. R. Mangun. 2009. Cognitive Neuroscience: The Biology of the
Mind. 3rd ed. New York: Norton.

Huguet, G., E. Ey, and T. Bourgeron. 2013. The genetic landscapes of autism spectrum disorders.
Annual Review of Genomics and Human Genetics 14 (1): 191–213.

Izuma, K., K. Matsumoto, C. F. Camerer, and R. Adolphs. 2011. Insensitivity to social reputation
in autism. Proceedings of the National Academy of Sciences of the United States of America 108 (42):

Jordan, H., J. E. Reiss, J. E. Hoffman, and B. Landau. 2002. Intact perception of biological
motion in the face of profound spatial deficits: Williams syndrome. Psychological Science 13 (2):

Koster-Hale, J., and R. Saxe. 2013. Functional neuroimaging of theory of mind. In Understanding
Other Minds, 3rd ed., ed. S. Baron-Cohen, M. Lombardo, and H. Tager-Flusberg, 132–163. Oxford
University Press.
Concept Nativism and Neural Plasticity 145

Koster-Hale, J., R. Saxe, J. Dungan, and L. L. Young. 2013. Decoding moral judgments from neural
representations of intentions. Proceedings of the National Academy of Sciences of the United States of
America 110 (14): 5648–5653.

Kovács, Á. M., E. Téglás, and A. D. Endress. 2010. The social sense: Susceptibility to others’ beliefs
in human infants and adults. Science 330 (6012): 1830–1834.

Kriegeskorte, N., M. Mur, D. A. Ruff, R. Kiani, J. Bodurka, H. Esteky, K. Tanaka, and P. A. Bandet-
tini. 2008. Matching categorical object representations in inferior temporal cortex of man and
monkey. Neuron 60 (6): 1126–1141.

Lakusta, L., B. Dessalegn, and B. Landau. 2010. Impaired geometric reorientation caused by
genetic defect. Proceedings of the National Academy of Sciences of the United States of America 107 (7):

Laurence, S., and E. Margolis. 2002. Radical concept nativism. Cognition 86: 25–55.

Laurence, S., and E. Margolis. Unpublished ms. Building Blocks of Thought.

Leslie, A. M., and L. Thaiss. 1992. Domain specificity in conceptual development: Neuropsycho-
logical evidence from autism. Cognition 43 (3): 225–251.

Lingnau, A., L. Strnad, C. He, S. Fabbri, Z. Han, Y. Bi, and A. Caramazza. 2014. Cross-modal plas-
ticity preserves functional specialization in posterior parietal cortex. Cerebral Cortex 24:541–549.

Lomber, S. G., M. A. Meredith, and A. Kral. 2010. Cross-modal plasticity in specific auditory
cortices underlies visual compensations in the deaf. Nature Neuroscience 13 (11): 1421–1427.

Luo, Y. 2011. Do 10-month-old infants understand others’ false beliefs? Cognition 121 (3):

Mahon, B. Z., S. Anzellotti, J. Schwarzbach, M. Zampini, and A. Caramazza. 2009. Category-

specific organization in the human brain does not require visual experience. Neuron 63 (3):

Mahon, B. Z., and A. Caramazza. 2009. Concepts and categories: A cognitive neuropsychological
perspective. Annual Review of Psychology 60 (1): 27–51.

Majewska, A. K., and M. Sur. 2006. Plasticity and specificity of cortical processing networks.
Trends in Neurosciences 29 (6): 323–329.

Marcus, G. 2004. The Birth of the Mind: How a Tiny Number of Genes Creates the Complexities of
Human Thought. New York: Basic Books.

Moran, J. M., L. L. Young, R. Saxe, S. M. Lee, D. O’Young, P. L. Mavros, and J. D. Gabrieli. 2011.
Impaired theory of mind for moral judgment in high-functioning autism. Proceedings of the
National Academy of Sciences of the United States of America 108 (7): 2688–2692.

Onishi, K. H., and R. Baillargeon. 2005. Do 15-month-old infants understand false beliefs? Science
146 Stephen Laurence and Eric Margolis

Pascual-Leone, A., and R. Hamilton. 2001. The metamodal organization of the brain. Progress in
Brain Research 134:427–445.

Peterson, C. C., J. L. Peterson, and J. Webb. 2000. Factors influencing the development of a
theory of mind in blind children. British Journal of Developmental Psychology 18 (3): 431–447.

Pinker, S. 2002. The Blank Slate: The Modern Denial of Human Nature. New York: Penguin.

Pinker, S. 2007. The Stuff of Thought. New York: Viking.

Polk, T. A., J. Park, M. R. Smith, and D. C. Park. 2007. Nature versus nurture in ventral visual
cortex: A functional magnetic resonance imaging study of twins. Journal of Neuroscience 27:

Prinz, J. 2005. The return of concept empiricism. In Handbook of Categorization in Cognitive Science,
ed. H. Cohen and C. Lefebvre, 679–695. Amsterdam; Boston: Elsevier Science.

Prinz, J. 2012. Beyond Human Nature: How Culture and Experience Shapes Our Lives. London: Allen

Quartz, S. R. 2003. Innateness and the brain. Biology and Philosophy 18 (1): 13–40.

Quine, W. V. O. 1969. Linguistics and philosophy. In Language and Philosophy: A Symposium, ed.
S. Hook, 95–98. New York: NYU Press.

Reiss, J. E., J. E. Hoffman, and B. Landau. 2005. Motion processing specialization in Williams
syndrome. Vision Research 45 (27): 3379–3390.

Renier, L. A., I. Anurova, A. G. De Volder, S. Carlson, J. VanMeter, and J. P. Rauschecker. 2010.

Preserved functional specialization for spatial processing in the middle occipital gyrus of the early
blind. Neuron 68 (1): 138–148.

Rezlescu, C., J. J. S. Barton, D. Pitcher, and B. Duchaine. 2014. Normal acquisition of expertise
with greebles in two cases of acquired prosopagnosia. Proceedings of the National Academy of Sci-
ences 111 (14): 5123–5128.

Rogers, T. T., J. Hocking, A. Mechelli, K. Patterson, and C. Price. 2005. Fusiform activation
to animals is driven by the process, not the stimulus. Journal of Cognitive Neuroscience 17 (3):

Rogers, T. T., and J. L. McClelland. 2004. Semantic Cognition: A Parallel Distributed Processing
Approach. Cambridge, MA: MIT Press.

Sadato, N., A. Pascual-Leone, J. Grafman, V. Ibañez, M. P. Deiber, G. Dold, and M. Hallett.

1996. Activation of the primary visual cortex by Braille reading in blind subjects. Nature 380:

Schubert, C. 2009. The genomic basis of the Williams-Beuren syndrome. Cellular and Molecular
Life Sciences 66 (7): 1178–1197.
Concept Nativism and Neural Plasticity 147

Senju, A., V. Southgate, Y. Miura, T. Matsui, T. Hasegawa, Y. Tojo, H. Osanai, and G. Csibra. 2010.
Absence of spontaneous action anticipation by false belief attribution in children with autism
spectrum disorder. Development and Psychopathology 22 (2): 353.

Senju, A., V. Southgate, S. White, and U. Frith. 2009. Mindblind eyes: An absence of spontaneous
theory of mind in Asperger syndrome. Science 325 (5942): 883–885.

Sharma, J., A. Angelucci, and M. Sur. 2000. Induction of visual orientation modules in auditory
cortex. Nature 404 (6780): 841–847.

Sripada, C. S., and S. Stich. 2006. A framework for the psychology of norms. In The Innate
Mind: Culture and Cognition, ed. P. Caruthers, S. Laurence, and S. Stich, 280–301. Oxford: Oxford
University Press.

Striem-Amit, E., O. Dakwar, L. Reich, and A. Amedi. 2012. The large-scale organization of “visual”
streams emerges without visual experience. Cerebral Cortex 22 (7): 1698–1709.

Sur, M. 1988. Visual plasticity in the auditory pathway: Visual inputs induced into auditory
thalamus and cortex illustrate principles of adaptive organization in sensory systems. In Dynamic
Interactions in Neural Networks: Models and Data, ed. M. A. Arbib and S.-i. Amari, 35–51. New York:

Surian, L., S. Caldi, and D. Sperber. 2007. Attribution of beliefs by 13-month-old infants.
Psychological Science 18 (7): 580–586.

Verhage, M., A. S. Maia, J. J. Plomp, A. B. Brussaard, J. H. Heeroma, H. Vermeer, R. F. Toonen,

et al. 2000. Synaptic assembly of the brain in the absence of neurotransmitter secretion. Science
287 (5454): 864–869.

von Melchner, L., S. L. Pallas, and M. Sur. 2000. Visual behaviour mediated by retinal projections
directed to the auditory pathway. Nature 404:871–876.

Warrington, E. K., and R. A. McCarthy. 1983. Category specific access dysphasia. Brain

Wilmer, J. B., L. Germine, C. F. Chabris, G. Chatterjee, M. Williams, E. Loken, K. Nakayama, and

B. Duchaine. 2010. Human face recognition ability is specific and highly heritable. Proceedings of
the National Academy of Sciences of the United States of America 107 (11): 5238–5241.

Wolbers, T., P. Zahorik, and N. A. Giudice. 2011. Decoding the direction of auditory motion in
blind humans. NeuroImage 56 (2): 681–687.

Young, L., and R. Saxe. 2009. Innocent intentions: A correlation between forgiveness for acciden-
tal harm and neural activity. Neuropsychologia 47 (10): 2065–2072.
III Concepts and Evolution
6 The Evolution of Conceptual Design

H. Clark Barrett

6.1 Introduction

The repertoire of concepts that a typical human adult possesses is immense. Depend-
ing on how you count them (a matter of debate), any of us easily has thousands
of concepts, and probably more like hundreds of thousands, or millions. There are
concepts that all of us probably share at least in some way, such as the concepts PERSON,
FOOD, MOTHER, WATER, and EYE. These concepts are not necessarily carbon-copy identical

in everyone who possesses them; an ophthalmologist’s concept EYE, for example,

may differ substantially from that of a nonexpert. Moreover, there may be substan-
tial diversity across individuals, cultures, times, and places in the concepts we possess.
Among the Shuar of Ecuador, many adults possess the concept ARUTAM, a spirit capable
of bestowing power on individuals under the hallucinatory effects of MAIKUA, a plant
in the nightshade family. Academic biologists possess concepts like INTRON, CLADE, and
ORTHOLOG. And our minds are populated by a vast diversity of idiosyncratic concepts of

people, places, events, objects, and practices, from BLACK SABBATH to BUNDT CAKE.
How are we to make sense of this? The concept CONCEPT, of course, is prescientific
in origin, and as such, there is no particular reason to expect it to pick out anything
real in the world, anything like a natural kind. Indeed, some have argued that the
concept CONCEPT is sufficiently muddled as to be not worth keeping (Machery 2009). But
as it happens, many scholars continue to use the concept CONCEPT in a scientific con-
text—particularly in philosophy, psychology, and neuroscience (Barsalou 1999, 2005;
Caramazza and Mahon 2003; Laurence and Margolis 1999; Carey 2009). If we want a
scientific theory of concepts, then, we will have to ask what concepts are, what they
do, how we come to have them, and why. And, given that humans are living organ-
isms shaped by the evolutionary process, these are fundamentally biological questions.
From an evolutionary point of view, why do organisms have concepts, what role do
they play in survival and reproduction, and how does natural selection shape the brain
machinery that causes us to have the conceptual repertoires that we, or any species for
that matter, come to possess?
152 H. Clark Barrett

One popular approach to understanding concepts from a biological point of view is

to start with innateness. Innateness has many definitions in the literature but gener-
ally refers to concepts or aspects of concepts that are not acquired via psychological
processes such as learning (Carey 2009; Samuels 2002; Spelke and Kinzler 2009).
Although innate concepts need not be present at birth, they are usually regarded as
remaining fixed or unchanging once they develop, and as developing the same way in
everyone, except for cases of developmental disorder or brain damage. Concepts that
are acquired or that change over the lifespan as a result of learning or other processes
of feedback from the environment or experience are therefore not regarded as innate—
or at least, not the altered forms they take after learning has occurred. The innateness
approach to concepts, then, attempts to ask what kernels of unlearned knowledge
are present in infancy that both continue through adulthood—often called core
knowledge—and that allow new concepts to be acquired (Carey 2009; Gopnik and
Meltzoff 1997).
Without denying the importance of the so-called starting state in infancy (or more
properly, in the zygote), here I would like to start from somewhere else: the developed
conceptual phenotype. From a biological point of view, the phenotype is what mat-
ters for fitness. This is because it is the phenotype that interacts with the world in the
service of survival and reproduction. What I mean by phenotype, in this case, is the
array of concepts that humans actually possess and deploy at a given point in their
lifespan—whether innate or learned—that make a difference in life, death, and repro-
duction at that moment. Thus, rather than focusing exclusively on the moment of
birth or the first year of infancy—or even on just those aspects of conceptual structure
that are identical across all individuals in adulthood—it is worth looking at the full
scope and diversity of concepts that humans come to possess, and to ask how this array
of concepts comes about because of natural selection acting on the developmental
systems that build our conceptual repertoires, both the parts that are universal and the
parts that vary. Importantly, this involves thinking about developed conceptual phe-
notypes as targets of selection: outcomes that natural selection has acted on, thereby
shaping the developmental systems that build them (Barrett 2006, 2007, 2015). As
a consequence, these developmental systems themselves have designs. The resulting
developmental designs are likely to be more complex than simply innate starting states
plus general-purpose learning rules, because it is the developmental outcomes, not just
the beginnings, that they have been selected to produce. Over evolutionary time this
leads to processes akin to guided learning toward designed targets: the full palette of
our mature conceptual repertoires, both universal and idiosyncratic.
This is what I will call the conceptual design approach. Here I am using the term
design in its evolutionary sense, as in the evolved design of wings, lungs, the immune
system, stereoscopic vision, and the evolved developmental systems that build them
(Carroll, Grenier, and Weatherbee 2005). The central premise of this approach is that
The Evolution of Conceptual Design 153

concepts and the psychological machinery that builds them have been designed, or
shaped, by the evolutionary process to carry out certain functions; they exist to help
us navigate and act in the world in the service of survival and reproduction, or fitness.
By taking an adaptationist, or engineering, approach to conceptual design—akin to the
approach taken by functional morphologists who study the design of animal limbs,
jaws, lungs, and other organs—we can ask, How do we expect natural selection to
shape the developmental machinery that allows us to acquire concepts, even concepts
that have a substantial or entirely learned component? Innateness may well be impor-
tant for thinking about such systems, but equally important will be thinking about the
design of developmental outcomes: conceptual phenotypes. Because these phenotypes
determine fitness—whether learned or innate—natural selection will shape whatever
heritable developmental resources influence their design.
In this chapter I make a case for the conceptual design approach and its poten-
tial for an evolutionary theory of concepts. To begin, I consider the question of what
concepts are, or what we want the term concept to mean, in an evolutionary sense.
Next, I present the conceptual design approach and the theoretical commitments it
entails. The bulk of the chapter then unpacks how this approach can be used to build
theories of conceptual architecture and the developmental machinery that shapes it,
using specific examples to illustrate the approach and focusing in particular on archi-
tecture organized around conceptual types. Finally, I consider the implications of the
conceptual design perspective for understanding our ability to acquire evolutionarily
novel concepts. My aim is to show how focusing on evolved design can orient the
study of concepts toward new directions that are consistent with the evolutionary
developmental or evo-devo turn in contemporary biology.

6.2 What Do We Want the Term Concept to Mean?

For a technical term to be of use, it is important for it to have a clear and unambigu-
ous referent. In the case of the concept CONCEPT, this has been a problem. Like many
concepts in the cognitive sciences, it has both folk and technical meanings (Stich
1983). As a result, there is not general agreement on what the concept CONCEPT means,
from a technical point of view, and diverse theories of concepts exist (see Laurence and
Margolis 1999; Machery 2009; Murphy 2002).
My intention here is neither to reject nor endorse any particular theory of what
concepts are, but rather, to begin with a working definition of concepts that is broad
enough to capture the range of possibilities for what concepts might be, while also
being specific enough to target those functions that most theorists generally hold
concepts to play in mental activity. In addition, our notion of concepts must be
biologically grounded, allowing us to ask what functions concepts play in the life
of organisms, and therefore, how evolutionary processes shape them. This, in turn,
154 H. Clark Barrett

implies two things we’d like in our theory of concepts. First, it should be broad enough
to include the concepts and conceptual machinery of nonhuman animals.1 Second,
it should be a functionalist notion of concepts—that is, a notion of concepts as infor-
mation structures that play a causal role in mental activity and behavior and not, for
example, a notion of concepts as abstracta that exist outside the mind (Margolis and
Laurence 2007).
Among the many functions, or roles, that concepts are held to play in mental life,
one can subsume most of them into two general classes of function, which I will
call generalization and inference (Murphy 2002).2 By generalization I am referring to
the grouping, or categorization, function of concepts: the concept COW, for example,
lumps together all cows for the purpose of conceptual analysis (this implies additional,
subsidiary functions, such as an identification function: group members must be iden-
tified as group members in order to group them). By inference I am referring to the many
informationally productive, or generative, things that can be done by grouping things
together conceptually. For example, one frequently discussed kind of inference enabled
by concepts is inductive inference, which occurs when I apply knowledge learned from
past encounters with cows to a new encounter with a particular cow that I have never
encountered before; for example, guessing that a newly encountered cow is able to
moo, even though I have not yet observed it doing so. Concepts also enable learn-
ing, and in particular, learning that is inferential. For example, if I observe a squirrel
making a predator alarm call and did not previously know that squirrels could make
such calls, I may now generalize this observation to all squirrels. I’ve thus learned
something about squirrels in general from observing a particular squirrel. Arguably,
this would not be possible—or at least, the learning process would be a different one—
if I did not have the concept SQUIRREL (see chapter 22).
For those familiar with the large literature on concepts, this reduction of conceptual
functions to two basic types might seem excessively simplistic.3 However, it satisfies

1. This is not meant to preclude the possibility that humans can have types of concepts that
other animals can’t or don’t have. Indeed, this is almost certainly true, and the type-based analy-
sis that I will be developing here allows for the possibility of conceptual types that are unique to
taxa. However, this approach does rule out the position, favored by some, that concepts are
unique to humans (e.g., that they are intrinsically language based). Although that might be a
defensible approach, it would restrict evolutionary analysis to a much smaller set of phenomena
than entertained here and would, among other things, preclude a comparative approach to the
evolution of concepts and conceptual design.
2. Technically, everything referred to below is a form of inference. These can be regarded as
subtypes that are often distinguished in the literature.
3. For example, it leaves out some frequently discussed functions of concepts, such as the role
they play in linguistic reference and communication. We will not preclude the possibility that
concepts may play additional roles, but functions that apply only to some subset of concepts
(e.g., lexicalized ones) should not be considered necessary features of concepts in general.
The Evolution of Conceptual Design 155

our basic objectives of capturing much if not most of what concepts are invoked to
explain in the psychological literature, while being sufficiently general as to not pre-
judge exactly how concepts might be instantiated or organized in animal minds, or
how they are distributed across species. Indeed, we would like those kinds of details
to emerge from an evolutionary theory of concepts, rather than being built into it. The
precise nature of the generalization and inference roles that particular concepts play is,
therefore, just what we want to explore from an evolutionary point of view.
A variety of basic questions are raised by an evolutionary perspective on concepts.
Perhaps the most basic are these two: Why do we have concepts at all, and why do we
have the ones that we do? To the extent that concepts and the mental machinery that
produces them are adaptations—the products of natural selection—the answers will
have to do with the precise nature of the functions that concepts play, which will entail
investigating the basic functions of generalization and inference in further detail. First,
however, let’s consider some specific examples of concepts, to get an idea of the scope
of the phenomena at hand.
The literature on concepts tends to focus on some well-worn examples. Perhaps the
most common are concepts of animal taxa, which include species concepts, such as
TIGER, and higher-level taxa, such as BIRD. Other common examples are FRUIT, VEGETABLE,

FURNITURE, and GAME—concepts whose properties call into question the classical view

of concepts as based on definitions, or sets of necessary and sufficient features (Rosch

1999; Smith and Medin 1981). There are also so-called natural kind concepts, which
include animal taxa like TIGER as well as naturally occurring categories of object or sub-
stance, such as WATER and GOLD (Keil 1992). Then there are concepts of individuals, such
as individual people (e.g., SHAKESPEARE, ALBERT EINSTEIN) and individual objects (THE MORNING
STAR, THE MOON, THE SPHINX). Already, this grab bag of concepts spans a broad range and

has spurred many debates about the nature of concepts. Perhaps most importantly, it
raises the question of whether concepts come in kinds or types. For example, are con-
cepts of natural kinds, like TIGER and GOLD, different from concepts of artifact kinds, like
FURNITURE? Are concepts of individuals, like SHAKESPEARE, different from concepts of groups

of things, like FRUIT? And what is the inventory of types of concepts the mind contains,
or can contain? The list above, for example, focuses heavily on physical objects and
substances (BIRD, VEGETABLE, GOLD), but clearly we can have non-object-based concepts as
well, such as our concepts JUSTICE, ANGER, and INTEGRAL.
How are concepts shared—or not—across species? Many basic concepts, such as
concepts of objects, causation, and agency, could very well be shared across species
(Barsalou 2005; Carey 2009). Some concepts clearly only humans have, or can have—
INTEGRAL, for example, or perhaps the concept FALSE BELIEF (Call and Tomasello 2008). And

concepts that are shared, at some level, might be different in the details. Although rats
appear to have an elementary understanding of causation, it’s not necessarily identical
to ours (Blaisdell et al. 2006). Tigers could and probably do have a concept of tigers,
which could share some features with our concept TIGER but presumably has others that
156 H. Clark Barrett

ours doesn’t. For example, the TIGER concept of tigers might include elements of smell,
or behavioral signals, that most human tokens of the concept TIGER don’t have. Within
species, perhaps especially in humans, there are likely to be individual differences
in the content of concepts, with biologists, tiger hunters, and four-year-old children
having TIGER concepts that probably overlap in some ways and differ in others. And it’s
likely that other animals have concepts that we don’t—for example, concepts used by
animals that navigate by the stars or magnetic fields.
A theory of conceptual design will need to be able to account for the repertoire of
concepts that species are able to acquire, as well as how these come to be distributed
with and across taxa via evolutionary processes of descent with modification. It will
also need to be able to account for the particulars of specific conceptual types, such
as the distinction between natural and artifact kinds—if, of course, these prove to be
legitimately different conceptual types in the human repertoire. And finally, it will
need to be able to account for things such as the creation and transmission of evolu-
tionarily novel concepts, such as INTEGRAL and QUARK, in species (notably us) that are able
to generate such novelties.

6.3 The Conceptual Design Approach

Natural selection shapes traits because of their effects on survival and reproduction
(fitness). Therefore, natural selection shapes concepts and the developmental machin-
ery that acquires them only (1) to the extent that concepts and conceptual machinery
are parts of organisms’ phenotypes and (2) to the extent that they play a role in help-
ing organisms survive and reproduce. Consideration (1) rules out the idea of concepts
as abstracta—we can only be considering concepts as they are actually instantiated in
neural tissue (whatever that instantiation might be). Consideration (2) suggests that
the criteria determining which conceptual abilities persist over evolutionary time must
be rooted in fitness and not some other currency such as truth value. This does not
mean that truth value cannot play a role in the evolution of concepts, but it does
mean that truth values that do not affect fitness are evolutionarily neutral and thus
not available to the designing force of natural selection. It also means that aspects of
concepts that are not true can be retained, if they have a positive effect on fitness. For
example, it is conceivable that concepts serving a fitness-promoting social role, such as
some religious concepts and social norms, might be favored by natural selection even
if certain aspects of their content are false.
However, the considerations above do not mean that we should leap straight from
fitness to concepts themselves. Natural selection shapes the developmental machinery
that allows us to acquire concepts but is not necessarily responsible for the details
of every conceptual token we possess, any more than natural selection acting on the
visual system is responsible for every detail of the specific visual representation of the
The Evolution of Conceptual Design 157

Figure 6.1
Reaction norms represent mapping functions between developmental environments and pheno-
typic outcomes (horizontal and vertical axes represent values of environmental and phenotypic
parameters, respectively). (A) A reaction norm in which phenotypic outcomes are a continuous,
smooth function of environmental state. (B) A step function in which the phenotype takes on
one of two discrete values depending on environment. (C) A canalized reaction norm: only one
phenotype develops, regardless of environment.

computer monitor I’m currently looking at. The machinery of conceptual development
evolved because the concepts it enabled our ancestors to acquire proved useful, but it
may also enable the acquisition of useless, or even fitness-detrimental, concepts (Tooby
and Cosmides 1990). Importantly, however, it is only by thinking about how natural
selection shaped the design of this machinery that we can build proper theories of the
conceptual repertoires that we are able to develop, even evolutionarily novel ones. This
idea—that evolution shapes developmental systems, which in turn build the concep-
tual phenotype that each of us comes to possess as adults—is the key to the conceptual
design approach (Barrett 2006, 2007).
Concept-building machinery can be seen as instantiating developmental input-
output functions, mapping between developmental inputs and phenotypic outcomes.
In biology these are known as reaction norms (Barrett 2012, 2015; Schlichting and Pigli-
ucci 1998) (see figure 6.1). Innate concepts are examples of flat (or canalized) reaction
norms, in which the same phenotype develops in everyone regardless of developmental
circumstance (barring extreme cases such as developmental disability).4 Domain-general
learning rules can be thought of as reaction norms in which the scope of phenotypic
traits affected is very broad. But many other kinds of reaction norms are possible as well.
To develop theories of the reaction norms that build conceptual phenotypes, we
must start by considering what functions concepts play, and therefore what form,

4. There is debate about whether canalization is equivalent to innateness as conceptualized by

most psychologists (Ariew 1996; Griffiths and Machery 2008). The point here is that develop-
mental universality can be seen as an example of a flat reaction norm.
158 H. Clark Barrett

or design, they can be expected to take. Ultimately, only by the signature that con-
cepts leave in behavior, and therefore in survival and reproduction, can natural selec-
tion shape them. The roles that concepts play in the mental pathways that lead to
behavioral decisions can be diverse and complex, however, involving the interplay of
many systems, including perception, learning, inference, judgment, decision making,
and motor control. Moreover, not all concepts or conceptual types need play the same
or similar roles in each of these systems, meaning that different conceptual types may
have very different design features.
Below I propose a tentative list of conceptual functions that can serve as a starting
point for thinking about conceptual design features. First, however, let me start with an
example of a conceptual type that can serve as a springboard for this discussion: FOOD.
There may be an overarching general concept FOOD that animals possess, but most
animals, including humans, are presumably also able to develop concepts of spe-
cific foods. In the terminology I will be using here, types and tokens are relative terms
referring to higher and lower levels in a hierarchy, respectively. Thus, food concepts
can be thought of as a type of concept (without, for the moment, prejudging the
issue of whether there has been selection for machinery specifically designed to acquire
this conceptual type). Within the general type, we can acquire specific tokens: FRUIT,
VEGETABLE, and MEAT might be tokens of food concepts nested within the general type

of FOOD. And there may be tokens nested within these as well: for example, APPLE and
ORANGE as tokens of the type FRUIT. Notice that this type/token terminology is relative

and does not necessarily privilege one level of the hierarchy as being the type. Some
scholars have suggested that there is a basic level in the conceptual hierarchy that is
most privileged or salient for the purposes of learning and inference (e.g., Rosch 1999).
This may indeed prove to be the case for some conceptual systems. Rather than assum-
ing it, we will leave this possibility open as one potential design feature that (some)
conceptual systems might possess.
Why do organisms have food concepts? Recall the two basic functions that I
mentioned above, generalization and inference. Let us assume for the moment that
squirrels have a concept NUT, that hummingbirds have a concept NECTAR, and that bears
have a concept BERRY.5 What good do these concepts do for these animals? In terms of

5. One could argue, again, that these animals do not have these “concepts” but have only behav-
ioral reactions to certain stimuli. From the functionalist perspective I’m adopting here, I’m not
considering this option to be distinct from the proposal that squirrels have a concept NUT, if squir-
rels generalize in some way across instances of the category of nuts (generalization), and if they
show common patterns of learning, inference, and behavior toward members of the category
(inference). Some accounts might want more for NUT to count as a concept, and indeed, some
species may have additional conceptual functions regarding nuts than squirrels do (e.g., a word
for nuts, the ability to talk about them). Here, we will just consider these to be additional possible
features or functions of concepts, which a given species might or might not have.
The Evolution of Conceptual Design 159

generalization, these concepts allow the animal to group objects (in this case) together
for the purposes of learning, inference, decision making, and behavior. This grouping
function can only work if there is an identification process—in this case perceptual,
though not necessarily for all concepts—that allows the squirrel, for example, to recog-
nize an individual nut as a nut. In most cases, the identification mechanism(s) must be
updatable by experience; for example, the squirrel might learn via tasting that a new,
perhaps perceptually distinct item has nutlike properties of taste and nutrition and
should be grouped with other nuts.
Once an item has been identified as a nut, this allows various conceptual resources
associated with nuts to be accessed—what we might call activating the concept NUT
(the nut is recognized as a nut, tagging the percept with the mental symbol for nuts).
Whether all the information associated with nuts should be regarded as part of the
concept is a question, perhaps semantic. For now we’ll entertain that concepts might
be complex, consisting of interacting information stores in the brain that might, in
some cases, be differentially activated in different circumstances (for example, not
everything that is part of your concept GEORGE WASHINGTON might be activated when you
see his face on a dollar bill, but we could still consider various facts that don’t come
immediately to mind as part of your GEORGE WASHINGTON concept).
The various inferential or informationally generative processes that occur when the
squirrel’s NUT concept is activated could be diverse. Let us use this example to develop
an initial list of possible conceptual functions, keeping in mind that this list might
not be exhaustive, and that different concepts might exhibit different combinations
of functions.

Learning The concept NUT could play a role in learning in several ways. Learned infor-
mation about nuts, such as their nutritional values, where to find them, and how
to handle them once encountered, could be stored as part of the NUT concept, to be
activated upon future encounters with nuts. Older information about nuts could also
be modified, that is, refinement of perceptual identification criteria upon encounter
with nutlike objects that are not nuts. Conceptual knowledge about nuts could also play
a role in learning about things related to nuts—for example, if we observe a raccoon
eating nuts, we might update our knowledge of raccoons to include a taste for nuts.
Inductive prediction Having a concept NUT allows knowledge learned from prior encoun-
ters with nuts to be applied to new encounters. For example, if a newly encountered
exemplar satisfies the criteria for being identified as a nut, the squirrel might assume
this new object is safe and even desirable to eat.
Reasoning Here I mean reasoning in the broadest sense to include various ways of com-
bining representations via rules to lead to new inferences. A squirrel that had object
constancy, for example, might infer that a nut that rolls into a hole still exists out of
sight (Baillargeon 1994; Carey 2009). This might prompt the squirrel to begin digging
160 H. Clark Barrett

to find the nut that it can no longer see. In humans, of course, concepts play a role in
many reasoning processes that probably do not exist in other animals.
Decision making Concepts would not exist if not for their role in decision making. For
example, knowledge of how nuts are distributed in the environment, and where they
can be found, can play a role in the squirrel’s foraging decisions. In optimal foraging
theory, information about the nutritional value of food items is combined with repre-
sentations of their distribution in the environment—how many, for example, are likely
to be encountered in a resource patch such as a tree—to shape animals’ decisions about
whether to continue foraging in the same patch or to switch patches (Charnov 1976;
Smith 1983).
Action Ultimately, decisions must manifest in action to have an effect on fitness.
Concepts can also play a role in how actions are executed. For example, knowledge
of how to access the nut inside of a nutshell can play a role in the squirrel’s choice of
actions to open the nut. Indeed, some concepts may be stored partly in the form of
motor representations, or have action representations associated with them (Barsalou
1999). In the case of some kinds of objects, such as human-made tools, action represen-
tations may be a key and possibly necessary part of the concept (Johnson-Frey 2004).
Communication One way in which concepts can play a role in communication is if they
are lexicalized, or in some way expressed symbolically through gestures or calls. For
example, many animals have specialized alarm calls that alert others to the presence
of predators, presumably by activating concept of a predator in others’ heads (Cheney
and Seyfarth 2007). More generally, signaling systems can entail activation of concepts
in the minds of others, such as mental state concepts related to aggression or mating,
or social role concepts like OPPONENT or TERRITORY OWNER.
Other interfaces Internal to the mind, concepts or conceptual symbols can also allow
different inferential systems to communicate with each other. Some scholars, for exam-
ple, propose that language serves as a medium for conceptual reasoning (Boroditsky
2001; Lakoff and Johnson 1980). It also seems likely that conceptual systems must be
able to interface with physiological systems such as, in the case of food, the appeti-
tive system (Jackendoff 1996). Being hungry makes us think about food and therefore
activates food concepts, putting us in motion to begin looking for actual tokens of food
in the world. This is presumably part of how the decision-making and action functions
of food concepts are instantiated, but it is important to note that conceptual systems
cannot exist in isolation from other body systems. They must interact or interface with
them to be of any use, and this in turn implies that there must be design features that
allow them to do so.

Again, this list of conceptual functions is not meant to be exhaustive. Other con-
ceptual functions can and surely do exist, and starting from a concept other than food
might generate a different list. However, we are now in a position to begin thinking
The Evolution of Conceptual Design 161

about how to develop hypotheses about the design of systems that populate the mind
with concepts.

6.4 Representational Formats

Let us start again with food concepts, expanding later to different domains. Empiri-
cally, we know that humans and other animals are able to acquire food concepts, and
we know something about what these concepts are like and what they do. The next
step is to begin formalizing the design features of food concepts in the language
of computational psychology, and then to think about how evolution shapes the
developmental systems that build them.
Computational theories of mind formalize mental processes as computational func-
tions operating over information structures, or representations (Marr 1982; Pylyshyn
1984). Although these computational functions, or mechanisms, must be instantiated
neurally, they could in principle take just about any computational form or input-
output relationship and include not just processes of inference and decision making
but also processes of learning and updating (Griffiths et al. 2010). Similarly, the
representations on which they operate are taken to be symbolic in the general sense
that they encode information, but many forms of representation are possible (e.g.,
representations may be distributed across different neural structures). Thus, such theo-
ries are intended to be descriptive at the computational level in Marr’s (1982) sense,
without foreclosing multiple possibilities for how the computational mechanisms and
representations might be instantiated neurally.
Where do concepts fit into such a scheme? Some approaches to concepts take
them to be forms of mental symbols and, therefore, representations on which
computational procedures operate (e.g., Pylyshyn 1984). It’s possible, however, that
some aspects of what we might take to be conceptual knowledge are instantiated
in the computational input-output functions themselves. For example, mechanisms
predicting the location of objects that have moved out of sight might embody, in
their computational procedures, principles of object behavior, such as rigidity, elas-
ticity, and responses to gravity and friction (Baillargeon 1994; Carey 2009). Thus,
in keeping with a broad-picture view of concepts that allows them to be complex
information structures with interacting parts, we’ll allow the possibility that con-
ceptual knowledge can be distributed across systems of representations and the
mechanisms that operate on them.
The design stance is particularly useful for thinking about the complementary
design features of evolved computational processes and the individual representations
they operate on. Among other things, it doesn’t partition innateness or learning purely
into one component or the other. Instead, we expect developmental systems to build
representations and computational mechanisms such that they interact in functionally
162 H. Clark Barrett

fitness-enhancing ways, and learning and other types of plasticity are tools that can be
used to do so. It is here that the notion of a reaction norm, with open parameters, or
slots that are variable yet formatted according to certain principles, becomes particu-
larly useful.
What computational format might food concepts take, and how might the
features of individual food representations be designed to interact with the computa-
tional properties of the brain processes that handle them? Imagine, for example, that I
observe a doughnut. This tokens an object representation in my visual system, and in
my visual working memory in particular. This representation, in the format that the
visual system uses, is then fed into various object recognition procedures, which, based
on the perceptual cues represented (shape, texture, color, and perhaps contextual infor-
mation such as my location in a doughnut shop), identify the object as a doughnut,
thereby activating my concept DOUGHNUT: a representation of a doughnut as a dough-
nut is tokened. A semantic tag (or its functional equivalent) is attached to the object
representation, such that my conceptual knowledge of doughnuts can be brought to
bear in guiding my inferences and decisions vis-à-vis the doughnut (for more on tags,
see Barrett 2005a, 2015).
Should we regard this particular object representation as a token of my concept
DOUGHNUT? Perhaps not. It is a token of a doughnut representation, to be sure, but we’ve

allowed that the information that is part of a concept writ large may be widely distrib-
uted. It might include, for example, motor routines for grasping doughnuts, recipes
for doughnuts, propositional knowledge of the history of doughnut making, episodes
of consumption of particularly memorable doughnuts, and the like. This might be a
disturbingly wide scope for DOUGHNUT to some, but there is no reason to foreclose the
possibility that conceptual systems orchestrate many cognitive resources.6
A useful question for thinking about conceptual design is, What features of
food concepts and their tokened representations are necessary, or at least typical, for
them to be able to interact appropriately with the computational systems that are
designed to use them? Of course, there might be many, and tokened representations
might be designed to interact with many systems. For example, the doughnut repre-
sentation must have whatever features are necessary for it to be handled by the object-
representation system, and for it to be fed into motor-planning systems that guide my

6. An atomic view of concepts would regard only the mental symbol or tag for doughnut to be the
concept DOUGHNUT, and all the knowledge and content I’m describing here would be external to
the concept itself (e.g., Fodor 1998). One could formulate such a theory in evolutionary terms,
limiting the function of concepts merely to this symbolic function in the language of thought.
On the present account, this is one function of concepts, but there is no principled reason not to
include other functional aspects of conceptual systems as part of an evolutionary theory of
conceptual design.
The Evolution of Conceptual Design 163

hand to the appropriate place when I reach for the doughnut, such as information
about the location of the object in space.
Of course, representations of object locations are not specific to food objects. This
demonstrates that a given token may in fact be a member of multiple representational
types (just as a token trout is also a token fish, vertebrate, animal, and object). What
computational properties might be specific to the doughnut representation as a repre-
sentation of food as food?
An obvious one is nutritional value. There is debate over exactly how nutritional
value is represented by animal food-representation systems—calories? protein? fat?
vitamins? separate represented values, or a composite value?—as well as what deci-
sion rules have evolved to guide animals’ decision making with respect to food (Smith
1983). It is not in doubt, however, that some kind of representation of nutritional value
is a necessary component of food concepts in order for animals to make decisions
about what to eat. And other represented values are almost certainly necessary for a
food decision-making system to work; for example, representations of where the food
type is located or likely to be found, cues to its likely presence (e.g., smell), and other
properties such as difficulty of processing or extracting the resource once found (think
about, for example, how you’d decide whether it’s worth it to try to get honey out of a
beehive when encountered) (Charnov 1976; Smith 1983). These, then, are among the
minimal or typical representational requirements for a token of a food concept.
What this means is that we would expect tokens of food concepts that develop in
adults to have a set of typical, or perhaps required, design features, which in this case
take the form of representational parameters, or slots that must be filled (Boyer 2002).
For example, food concepts might possess a caloric value slot (figure 6.2).7 Although it’s
an open question just what information the brain stores about the nutritional value
of foods, for food concepts to play their fitness-relevant role in eating decisions, there
must be nutritional value parameters, such as caloric content, that are either present
or waiting to be filled in each token of food concepts that we possess. Similarly, there
must be another parameter or slot for taste, smell, appearance, and probably more.8
Beyond design features of representations themselves, there are the design features
of computational systems that take those representations as inputs. Representational
formats must be designed to interface with the computational processes that use them,
and vice versa. If food concepts and their tokens have nutritional value slots, systems

7. One could imagine impoverished or incomplete food concepts in which no nutritional

value was represented, but presumably this means an empty slot in the token of the concept.
Empty slots might mean the conceptual token cannot be used for certain kinds of decision
making about food.
8. In Boyer’s (2002) terms, these are slots in conceptual templates. The way I am using conceptual
types here is analogous to conceptual templates.
164 H. Clark Barrett

Figure 6.2
Example of a conceptual type (food) instantiated in multiple conceptual tokens. Individual
parameters are open in the type (square brackets) and specified in the tokens by environmental
input, via developmental reaction norms.

for evaluating and making decisions about food must be designed to expect informa-
tion in those slots and have computational procedures that use the information in
principled ways. For example, decision rules in optimal foraging theory take represen-
tations of food density in a patch, coupled with representations of the nutritional value
of each item, as inputs and deliver decisions about how long to forage in the patch,
using a specific computational rule (Charnov 1976). Similarly, systems for learning
about food must be designed to systematically update the class-level representation of a
particular food type—such as doughnuts—as a function of experiences with individual
tokens of the type, amassed over time; for example, such systems could instantiate
Bayesian updating algorithms (Tenenbaum, Griffiths, and Kemp 2006). But for such
learning procedures to work, they would need specific priors about the parameters in
question—for example, how much to update the summary representation of doughnut
nutrition as a function of eating a specific doughnut. And there must be other designed
interfaces as well, as in the interface that allows us to inform others that a doughnut
is present when we visually observe a doughnut—an interface between perceptual,
conceptual, and linguistic systems (Jackendoff 1996).
What is the taxonomy of conceptual types that minds possess or can possess,
and how can we answer questions about the grain of domain specificity of the types
The Evolution of Conceptual Design 165

(Atkinson and Wheeler 2004; Barrett 2009)? In other words, how might we know if
there is conceptual machinery specific to foods and food concepts, or if this machinery
is just a token of some more general type, with no dedicated evolved machinery for
foods per se? For example, what if there is just a single, domain-general concept-
formation system, capable of representing any of the property types that a given
species’ conceptual capacities can represent?
This is of course possible, and a combination of theory and empirics is necessary
to answer the question. We’ll return to the question of domain taxonomy below, but
first consider two points. First, it is a feature of taxonomies that generality at one
level doesn’t necessarily preclude greater specificity at another level. For example, our
concept OBJECT and the conceptual resources for thinking about objects may be both
quite broad (in terms of the scope of objects that can be conceptualized) and evolu-
tionarily specialized (e.g., there may exist evolved systems specifically for representing
and thinking about objects; Carey 2009). But this does not preclude the possibility of
additional, specialized systems for specific categories of objects such as foods, faces, or
artifacts (Boyer 2002; Kanwisher 2010). In other words, it’s possible if not likely that
objects and foods are not separate domains but either nested or partially overlapping
ones. Our taxonomy of conceptual types will need to consider this kind of possibility.
Second, the idea of designed interfaces between conceptual and other systems sug-
gests that one all-purpose conceptual system might not do the trick. It’s likely, for
example, that food concepts have specialized and ancient interfaces with systems such
as appetite and digestion, required to extract nutritional information and to adjust
motivations for food seeking as a function of nutritional state. A general-purpose
conceptual system would need to reengineer the links between food concepts, food
learning, and these bodily systems each generation—possible, but perhaps less plau-
sible than a designed interface, suggesting at least some domain specificity in food
Other categories of concept, on the other hand, might require different represen-
tational formats and designed interfaces of their own, ones that are distinct from or
only partially overlap with food. We’ve already seen the case of objects, whose concep-
tual machinery needs to interact in principled ways with visual and motor systems,
and which pose a different (though again, partly overlapping) set of representational
demands. As an additional example, consider the set of concepts used in mentalistic
reasoning or theory of mind, such as the concept BELIEF. The format of belief representa-
tions is presumably quite different from the format of food representations, with non-
overlapping conceptual slots. For example, belief representations represent something
like attitudes toward states of affairs in the world (Leslie 1994). This entails informa-
tional slots that are clearly not present in food representations. Foods are not repre-
sented as having attitudes or standing in referential relationships to states of affairs.
Moreover, the slots and parameters of belief representations are presumably designed
166 H. Clark Barrett

to interface with an inferential apparatus designed to predict and retrodict the behavior
of other agents on the basis of their represented beliefs—a kind of causal reasoning that
is distinct from decision making about food in several ways (foods, for example, are
not conceptualized as intentional agents whose external behavior is caused by inner
representational states).
Although these considerations aren’t enough to settle the issue of whether there are
separate domains or subdomains for food and mental-state concepts, it suggests the
kinds of functional-specificity considerations we’ll need in order to develop properly
testable theories of conceptual types. Let us now turn to the issue of developmental
systems, and then return to the issue of conceptual repertoires.

6.5 Developmental Systems

Above I introduced the biological concept of a reaction norm (figure 6.1). In biology,
this is typically viewed as a mapping function between some environmental param-
eter space and a phenotypic parameter space, for a given genotype (Schlichting and
Pigliucci 1998). This concept can be broadened so that the phenotypic parameter space
includes mental phenotypes, such as representations and computational procedures for
operating on them (Barrett 2012, 2015). The environment, in turn, can be expanded to
include various aspects of an organism’s informational environment, including, in the
case of humans, our cultural and linguistic environments, as well as social cues such as
gestures, facial expressions, and the like.
In this framework, the developmental systems that build conceptual machinery,
including conceptual tokens, are reaction norms shaped by selection as a function of
the fitness value of the conceptual phenotypes they produce. Just as in the computa-
tional theory of mind, these reaction norms are descriptions of developmental systems
at the computational level, leaving open various possibilities for how they might be
instantiated at the developmental hardware level (i.e., genes, epigenetic machinery,
and other developmental resources). One can imagine several kinds of environment-to-
phenotype mapping functions these reaction norms might instantiate.
Most obviously, there must exist reaction norms that fill the open informational
parameters of evolved representational types and computational procedures. The
input-output rules of reaction norms determine how environmental experience is used
to populate these open parameter spaces. In biology it is typical to represent these
graphically as one-to-one functions, or curves, where each value of some environmen-
tal parameter is mapped to a value of the phenotypic parameter; an example might
be the function that maps amount of childhood nutrition to adult height, for a given
genotype. In psychology, it is relatively straightforward to imagine various learning
rules or algorithms instantiating environment-to-phenotype mapping functions. In
the behaviorist tradition, for example, the reinforcement schedule would provide
The Evolution of Conceptual Design 167

the input, and the developed behavioral phenotype would be the output, with con-
ditioning laws as the reaction norm (Gottlieb 2006). In a Bayesian framework, algo-
rithms specify a computational pathway from informational inputs to representational
outputs—including, in recent models, conceptual structures—given the appropriate set
of priors (including, perhaps, innate priors; Tenenbaum, Griffiths, and Kemp 2006).
In principle, any kind of mapping function could be instantiated as a reaction
norm—provided, of course, that a variant of the appropriate developmental system
becomes available for natural selection to act on. This might include not just curve-
like reaction norms that map from a one-dimensional environmental parameter (e.g.,
average experienced caloric value of a food item) to a one-dimensional representational
parameter (represented nutritional value of the food), but other more complex types
of function as well. One can imagine a reaction norm that maps experience with a
given lexical item being used referentially, for example, cow, to an adult phenotype in
which the word cow activates the concept COW. Arguably, this is a reaction norm that
has been shaped by natural selection—but if so, it is a complex one that could not
easily be represented on a graph mapping one linear dimension (environment) onto
another (phenotype). Instead, the input and output spaces are of high dimensionality.
Similarly, Chomsky’s notion of a language-acquisition device could be seen as instan-
tiating a complex reaction norm that maps the set of experienced linguistic utterances
to a developing conceptual structure of the language’s grammar (Chomsky 1965).
Our task in describing concept-building systems is to properly describe the underly-
ing reaction norms in computational terms, and to theorize how they are shaped by
One way of thinking about reaction norms is as spawners of phenotypes: they popu-
late the mind with its complement of mechanisms and representations (Barrett 2012,
2015). It is easy to imagine developmental reaction norms spawning individual tokens
of concepts and other representations; for example, my concept ELECTRON was brought
into being in my brain, from nothing, by some (presumably quite complicated) path-
way of developmental procedures. And the design approach suggests that we may also
think of mental mechanisms themselves as being spawned during development by
procedures using some combination of innate structure plus experience.
Innateness can be accounted for in reaction norm models. For example, aspects
of phenotype that develop the same way in everyone, regardless of environment, are
flat, or canalized (Ariew 1996; though again, canalization and innateness might not
be identical on some definitions; Griffiths and Machery 2008). Adding a temporal
dimension to the graph would capture at what age such canalized features develop. But
importantly, flatness might be a property of some dimensions or aspects of a reaction
norm, but not others. For example, all my object concepts, from APPLE to ZEBRA, might
have some elements in common, and we could conceptualize this as a dimension of
the object-concept phenotype along which the reaction norm is flat. But clearly, my
168 H. Clark Barrett

concepts APPLE and ZEBRA also differ in many ways (e.g., represented shape and color; one
represented as a food, the other as an intentional agent; etc.). These would be nonflat
dimensions of the concept-spawning reaction norm: parameters that are filled using
some combination of developmental rules and environmental inputs.
Crucially, some aspects of reaction norms that appear flat across the range of
environments humans experience might not be flat if a child were raised outside that
environmental range. For example, some aspects of the concept GRAVITY, including rules
for predicting how objects will fall or not fall depending on how they are supported,
might develop the same in all children (Baillargeon 1994). If what we mean by innate-
ness is a flat reaction norm across the environmental range in question, then these
aspects of the concept GRAVITY might appear to be innate, if they develop similarly in
most children. However, even flat reaction norms do not imply that learning can’t be,
or isn’t, involved in their construction. Indeed, it could well be that children every-
where need the appropriate experience to properly develop concepts such as GRAVITY,
SOLIDITY, COLLISION, and the like—but that all children on earth get the same relevant

experience (Baillargeon 1994; Barrett 2006). It is an open question whether a child

raised in space would develop the same concept of gravity as a child raised on earth.
What rules or computational procedures do reaction norms use to fill the open
parameters of conceptual phenotypes? Again, the possibilities are many, constrained
only by what kind of developmental machinery can be built out of genes, neurons,
and other biological building blocks (Gallistel and King 2009). However, the design
approach can be used to generate hypotheses about the minimal set of features a given
reaction norm must have—a logic akin to Chomsky’s learnability approach to specify-
ing the minimal parameters of a language-acquisition device (Chomsky 1965; Pinker
1989). For example, the process of fast mapping of lexical terms to concepts (Heibeck
and Markman 1987), in which children learn to map a word to a meaning in a single
trial, suggests, at minimum, a reaction norm that (1) spawns an auditory representa-
tion of the newly heard word in long-term memory and (2) associates it with a concept.
The reaction norm may well have other computational properties, but it must at least
have these. Similarly, creation of a new animal concept must entail spawning of a
new conceptual token and filling at least some of the slots required for identification,
for example, some minimal information about what the animal looks like (Barrett
and Broesch 2012). Other conceptual types, such as artifacts, might require additional
minimal information to spawn a new concept—for example, information about func-
tion might be necessary to spawn a new concept of an object as a tool (Kemler Nelson
et al. 2003). In each case, the shape of the relevant reaction norm will depend on the
nature of the conceptual phenotype it was designed to spawn, with some being quite
general and others more specific.
A final point about spawning is that it is presumably a normal part of the func-
tion of conceptual reaction norms to spawn novel tokens of conceptual types. Humans
The Evolution of Conceptual Design 169

undoubtedly do not have an innate concept CAKE, and yet, the concept CAKE is probably
a perfectly normal exemplar of both an object concept and a food concept. Similarly,
SCHADENFREUDE is perhaps not an innate mental-state concept (though it could be). Yet

the spawning of the concept SCHADENFREUDE given the appropriate experiential inputs
is presumably not outside the normal envelope of the mental-state concept-spawning
reaction norm. The token itself might be evolutionarily novel, but in each case, the
ability to spawn tokens of this type is not novel.
This is an important possibility that is often overlooked in discussions of the ability
of evolved systems to handle evolutionary novelty, because many scholars would hold
that the ability to acquire the concept CAKE—which is surely an evolutionarily novel
token—must be handled by so-called general purpose machinery (e.g., Chiappe and
MacDonald 2005). On the account here, the developmental machinery might in fact
be specialized, and spawning novel tokens could be exactly what the machinery was
designed to do. Systems for face recognition, for example, are probably evolutionarily
specialized (Kanwisher 2010), and yet every specific face that we see, every token, is
evolutionarily unique. Of course, this means that the interesting questions have to do
with what kinds of evolved conceptual types there are—and, relatedly, whether we have
any concepts that are not tokens, albeit novel ones, of evolved types. We now turn to
the question of conceptual types and their taxonomies.

6.6 Conceptual Types

On the theory outlined here, conceptual types are delineated by reaction norms:
a conceptual type exists when there is a reaction norm designed by natural selection
to produce tokens of that type. Importantly, conceptual types can be hierarchically
organized. Just as muscle cells are a type, and striated and smooth muscle cells are
subtypes within it, one can imagine the possibility of subtypes of concepts within,
for example, an overarching category of object concepts, as well as crosscutting types
(Barrett 2012).9 In the limit, there may be reaction norms designed to produce a single
token of a concept; Konrad Lorenz’s geese, for example, might have had just a single
concept MOTHER, attached to only one token in the world, by design.
If reaction norms are designed to produce tokens of conceptual types, then they are,
by definition, adaptations (i.e., designed by selection). As for any adaptations, the grain
of the adaptation—how broad or narrow its function or domain is—depends on facts
of both descent (evolutionary history) and modification (selection pressure). For exam-
ple, it’s possible to imagine a species with only a single concept-formation mechanism

9. What I mean by crosscutting is that some subtypes of a type might have elements or properties
that extend outside the domain of the type in which they are nested. Food concepts, for example,
link to nonspatial representational systems having to do with appetite, hunger, and bodily state.
170 H. Clark Barrett

because no variants of that mechanism ever appeared for selection to act on (a fact
of evolutionary history). If and when variants appear, whether there will be selection
for two distinct mechanisms—as opposed to the new variant failing to be selected
for, and disappearing—depends on whether there is selection for two, as opposed to
one, mechanism. This will depend on facts about what the reaction norms do, and
the relative costs and benefits of two versus one—considerations related to the notion
of functional incompatibility, or whether subdividing a function results in higher
fitness than maintaining a single functional design (e.g., Hughes 1994; Sherry and
Schachter 1987).
How, then, do we go about determining the catalog of conceptual types that a
species, such as humans, possesses? Clearly it’s ultimately an empirical matter, and
there do exist proposals about the array of conceptual types. For example, Carey (2009)
proposes that humans possess several core concepts or conceptual systems (OBJECT,
NUMBER, AGENT, CAUSE), as well as domain-general learning mechanisms, including

Quinian bootstrapping, which together produce all the conceptual tokens that adult
humans come to possess. Although this is a cogent proposal with substantial evidence
to support it, there are other possibilities.
For example, it is possible that objects, number, agency, and causality represent
aspects or dimensions of reaction norms—perhaps even flat, or canalized, dimen-
sions—which build tokens of these concepts that, while uniform along these dimen-
sions, may be variable along others. Humans could possess multiple kinds of object
concept, as reviewed above, with different parametric slots in addition to the core ones
(e.g., a horse is an object that is both a natural kind and an agent; a bench is an object
with neither of those properties). And there might be multiple kinds of agents, again
defined by properties additional to the core ones. This is not necessarily contradictory
to Carey’s proposal in some ways, but could diverge in others. For example, it’s possible
that there is no token of the general abstract concept OBJECT or CAUSE that is innate.
Moreover, it’s possible that learning is required to construct the developed versions
of any of these concepts, and it’s possible that at least some of the learning involves
mechanisms specific to the conceptual type, rather than being purely domain general.
It’s also possible, if not probable, that the minds of humans and other animals con-
tain conceptual types in addition to the ones Carey proposes. Foods are one type that
I have discussed at length. Foods are certainly represented as objects (and some may
be a special kind of object, sometimes known as a stuff or substance, which can have
different physical dynamics than other kinds of object; Prasada, Ferenz, and Haskell
2002). But as reviewed above, they almost certainly entail additional specialized devel-
opmental machinery, probably evolutionarily ancient, that not all objects share (this
is what I mean by crosscutting as opposed to entirely hierarchical concepts). Other
likely evolved conceptual types, probably ancient in mammals, include kinship (and
perhaps particular kin types such as PARENT and OFFSPRING), territory, taxa (CONSPECIFIC,
The Evolution of Conceptual Design 171

HETEROSPECIFIC) and biological roles: PREDATOR, PREY, and MATE (Barrett 2005b). Clearly

these are related to concepts of objects and agency, but it is not clear that they can be
subsumed within these core systems or simply emerge from the properties of those
systems without additional evolved machinery. And humans may have unique con-
ceptual types, such as concepts of tools, social norms, and moral principles, that
again could potentially emerge simply from these other core systems (i.e., without
additional selection), or could alternately be supported by machinery specifically
evolved to generate those types. A priori preference won’t tell you which it is; you need
empirical tests.
How do we know whether any proposed taxonomy of conceptual types is the
correct one? At present, the empirical study of concepts is poorly suited to adjudi-
cating between models. As elsewhere in psychology, mutually inconsistent theories
of concepts and conceptual structure are able to coexist in the literature for long
periods. Potentially, new methods such as model comparison hold promise. For
example, statistical model-comparison techniques allow comparison of models with
different levels of specificity or generality by controlling for the ability of models with
more free parameters to better fit data (Burnham and Anderson 2002). In principle,
this could allow a model with four conceptual types to be compared with a model
that has more, by seeing which better fits the data in tasks such as categorization and
learning tasks. Kemp and Tenenbaum use hierarchical Bayesian models to evaluate
whether empirically measured patterns of inductive inference are more consistent with
a hierarchical or nonhierarchical conceptual structure, a powerful statistical tool for
uncovering the existence of multiple conceptual types (Kemp and Tenenbaum 2008;
Tenenbaum, Griffiths, and Kemp 2011).
My goal here is not to engage in empirical evaluation of any particular proposal of
conceptual taxonomy, but rather, to offer a new theoretical framework for thinking
about conceptual types and conceptual design. In my view, it seems likely that the
mind contains a larger taxonomy of conceptual types, some evolutionarily ancient and
some recent, than is generally recognized in the literature. Part of this lack of recogni-
tion may be due to the focus on innateness, since many designed conceptual types
may exist only as (partly) learned tokens. Another part may be due to the tendency of
psychologists to focus on humans, without considering the large scope of concepts that
humans may share with other animals, as well as the unique concepts that humans are
able to form that must, on evolutionary grounds, require human-specific adaptations
to form (otherwise, other species would be able to do so).
Yet another part of the reason for the relatively impoverished view of conceptual
types that exists in the literature may be due to how domains are typically conceptu-
alized: as mutually exclusive rather than potentially overlapping and combinatorial.
On the view proposed here, it’s possible that conceptual tokens may in fact exist in
multiple domains at once, meaning that different aspects of these conceptual tokens
172 H. Clark Barrett

are assembled by different reaction norms. If this is the case, then some seemingly
opposed domain-general and domain-specific accounts of concept formation and
acquisition might be simultaneously true. For example, concepts might be formed by
Bayesian learning procedures, and there may be specific parameters for specific concep-
tual types. If true, the space of possible conceptual types might represent not mutually
exclusive domain sets, but rather an intersection of the sets of conceptual types made
possible by the brain’s diverse reaction norms.
Finally, this raises a point about the phylogenetic distribution of conceptual types
across taxa. If conceptual types evolve through processes of descent with modification,
then there are likely to be some conceptual types that are both ancient and widely
shared—albeit modified in particular lineages, just as gene families and phenotypic
traits, like limbs, can be widespread but modified in different descendant lineages.
Conceptual types such as objects, food, animals, mates, kinship, and causation could
be ancient and widely distributed. Reaction norms for constructing conceptual types—
even very general ones such as Bayesian updating procedures or statistical learning—
could similarly be ancient and serve as building blocks to be mixed and matched in the
development of particular tokens. But this is not necessarily to say that, for example,
our concepts of animal taxa are in every way the same as those of other species. As
mentioned above, tigers almost certainly have a concept TIGER, and though the machin-
ery that builds that concept may be evolutionarily homologous to ours (i.e., descended
from a common ancestor), there is no reason that our concepts of tigers can’t be
different—and perhaps richer—through interaction with developmental machinery
that tigers don’t possess. For example, our TIGER concepts can include the lexical tag
tiger and propositionally stored information from scientific biology on the nature of
tigers that is not present in tigers’ own concept of their species.
In other words, processes of descent with modification of concept-building reaction
norms imply that concepts come in versions that may exhibit evolutionary similarities
and differences across taxa. And some species might possess whole classes of concepts,
descended from more ancient conceptual types, that others don’t possess. If true,
our capacity to form certain kinds of taxonomically unique concepts, such as the
concepts used in mathematics and biology, might have appeared not entirely de novo,
but rather via modification of evolutionarily older conceptual machinery in the lineage
leading to us.

6.7 Novelty, Flexibility, and Culture

Humans are clearly able to form new concepts: concepts that presumably were not
formed in our evolutionary past (and in some cases still might not be formed in all
humans). For example, until recently nobody had the concept MOBILE PHONE. Now, it’s
questionable whether there is anyone who doesn’t. There doesn’t seem to be anything
The Evolution of Conceptual Design 173

terribly odd or unusual about a concept of mobile phones; many cultures, for example,
have concepts of objects that can cause things at a distance. But are there concepts that
challenge the view that every token concept we can form is a token of some evolution-
arily older type? Carey (2009) gives as examples the concepts ELECTRON, CANCER, INFINITY,
and GALAXY. How are we to account for these?
A widely held solution to the problem of novelty is the existence of domain-general
learning abilities that have not been selected for any particular phenotypic outcome.
However, the fact that concepts like CAKE and MOBILE PHONE don’t seem particularly
outside the envelope of normal human concepts should give us pause, because these
certainly are evolutionarily novel conceptual tokens. Although cake presumably didn’t
exist in human ancestral environments, the concept CAKE seems to be a fairly ordinary
case of a human-made food prepared from (in theory) naturally available ingredients:
a food-artifact hybrid. Similarly, mobile phones are evolutionarily novel, but the
concept MOBILE PHONE is of an object, an artifact, with the interesting causal property of
being able to transmit the voice over a distance. It’s not hard to imagine this concept
being assembled from conceptual resources present long before mobile phones existed
(objects, artifacts, causation-at-a-distance, which is a normal feature of communica-
tion). Thus, the capacity to generate phenotypic novelty does not necessarily require
domain-general mechanisms, or imply that novel conceptual tokens can’t be tokens of
evolutionarily non-novel types.
Are concepts such as ELECTRON and INFINITY outside the scope of this model? Not neces-
sarily. Even on accounts such as Carey’s, which depends on domain-general learning,
these concepts are assembled at least in part via modifications of innate concepts. Most
people, for example, conceptualize electrons as objects (at least when first learned),
and infinity is a concept arrived at by imagining the limitless extension of a counting
process, which on Carey’s account makes use of innate resources. On the account here,
the ELECTRON concept would be (potentially) a learned token of an OBJECT concept. It
would have some interesting and unusual properties, such as extremely small size—but
there is no reason to think that property is unrepresentable as a value of an evolved
parameter for object concepts, that is, size (note that contemporary physics consid-
ers electrons to be point particles with zero size, and those who fully understand the
concept may genuinely be able to grasp this; but it is plausible that for most learners
the initial concept is a token of the concept OBJECT, and even physicists may recruit con-
ceptual machinery for thinking about objects when thinking about electrons).
Are there evolutionarily unrepresentable parameters, parameters that require, in
Carey’s terms, construction of entirely new conceptual resources? It’s possible, but it
depends on what one means by this language. For example, it’s clearly the case that
whatever evolved conceptual resources we have, including our evolved reaction norms,
make possible the development of a concept such as INFINITY (arguably, as Carey suggests,
a linguistically represented concept grounded in a statement about recursion that is,
174 H. Clark Barrett

itself, not represented any more deeply). But a different question is whether genuinely
new representational formats, ones that have not been selected for, are required to form
concepts such as ELECTRON and INFINITY. Whatever is at stake here lies in hypotheses about
what the representational formats required are, and what the preexisting repertoire
of possible formats is. My suspicion is that most if not all novel concepts make use of
previously available representational formats—a form of neural reuse (Anderson, 2010,
Barrett, 2012; Dehaene, 2009)—but this is an open question that cannot be resolved in
the absence of more specific theories of conceptual types.
This idea of novel tokens of evolutionarily preexisting conceptual types does not
mean that humans don’t have evolutionarily unique conceptual capacities. It just
means that these have been selected for and therefore have functions and design
features. For example, as Carey and others argue, spoken language probably makes
possible the formation, transmission, acquisition, and representation of concepts that
other species can’t acquire. Many mathematical concepts, for example, may be depen-
dent on language and make use of its unique ability to combine linguistically encoded
symbols using explicit rules in a generative manner (INFINITY would be an example).
Also extremely important in humans are processes of cultural transmission (including
language) and, of perhaps even greater importance, cultural evolution, which allows
the accumulation of conceptual complexity in areas such as mathematics and science
(Richerson and Boyd 2005).
If this perspective is correct, then we should not be thinking about two channels
of concept acquisition, the innate channel (for evolutionarily old concepts) and the
domain-general linguistic and cultural channel (for evolutionarily novel concepts).
Instead we should be asking how older conceptual reaction norms interact with newer
mechanisms and processes, including language and cultural evolution. Indeed, if
spoken language and cumulative cultural evolution are unique to humans, then
these skills evolved from their very beginning against a developmental background of
existing conceptual reaction norms. We should therefore be asking how these new
capacities evolved to exploit the older ones, as much as the other way around. Part of
this will involve the study of designed interfaces between older and newer systems. For
example, how does our ability to talk and propositionally reason about food—new—
develop on top of a much older conceptual apparatus for developing food concepts and
linking them to bodily systems?
A useful case study for this is work investigating how written words—a paradigm
case of culturally evolved symbolic artifacts—are represented and processed in the brain
(Cohen and Dehaene 2004; Dehaene 2009). In a nutshell, it appears that exposure to
written language in childhood causes the spawning of an evolutionarily novel token of
an object-recognition module, an evolutionarily old type, in the brain’s visual cortex.
This module shares common features across readers of all the world’s languages but is
absent in nonreaders—just as one would expect from a contingent, yet functionally
The Evolution of Conceptual Design 175

specialized, reaction norm (Dehaene 2009). Interestingly, Changizi and colleagues

have found evidence that processes of cultural evolution have shaped the world’s
written characters to make them more easily processed by the brain’s object-
recognition procedures, suggesting that the properties of object-recognition modules
act as a selective filter in the process of cultural evolution of writing systems (Changizi
et al. 2006).
One can argue about whether this system instantiates anything conceptual, but it
serves as a kind of model for how evolved conceptual reaction norms might interact
with processes of cultural evolution of concepts. Just as the brain’s ability to discrimi-
nate objects acts as a selective filter on the evolution of characters, the brain’s ability
to acquire concepts can (and presumably must) act as a filter on cultural evolution—
—a point made in epidemiological or cultural attractor accounts of cultural transmission
(Boyer 2002; Sperber 1996). This is not to say that it filters along every parameter of
cultural possibility space; but it certainly filters some (if we were unable to acquire a
concept, how could it be transmitted?). And this filtering process undoubtedly occurs
in many other ways as well—for example, the evolution of tools is shaped by the
possibilities of human bodies to use them.
In the brain, then, we might expect a host of evolutionarily novel concepts, and
for our repertoire of concepts to be changing all the time—especially as the speed of
cultural change accelerates. But this should not be regarded as a process independent of
evolved conceptual machinery. Instead, the interesting questions for future research lie
in uncovering the parameter space of human conceptual repertoires, and asking how
these interact with processes such as cultural evolution.

6.8 Conclusion: Toward Understanding Normal Human Conceptual Diversity

One of the primary virtues of a conceptual design approach is that it does not view
the diversity and richness of adult conceptual repertoires as emerging from mere
happenstance, solely the product of identical innate starting states meeting a haphaz-
ardly patterned world. There is likely to be a reason, for example, that hunters grow-
ing up in foraging societies come to possess an elaborate conceptual understanding of
animal behavior and biology—and the proposal that this results from combining a few
innate conceptual primitives like CAUSE and AGENT, along with general-purpose learning
rules, might well be too impoverished to explain it (Liebenberg 1990). Instead, the
fleshed-out tokens present in the minds of adults, such as the richly detailed informa-
tion stored under a Shuar adult’s concept JAGUAR or TAPIR, could be the normal type of
developmental outcome that selection has shaped developmental systems to build—
even if the richness of these concepts is undetectable at birth or early childhood (Boyer
2002; Barrett 2005b). If so, then we will need to resist the simple equation of what
is present in infancy with what natural selection builds, and confront the richness
176 H. Clark Barrett

of adult concepts as what needs to be explained (without forgetting, of course, that

infancy is a crucial step in this process).
How do we do this, empirically? Because the picture of mental development implied
by a conceptual design view is substantially more complex than a small number of
innate concepts plus learning, the task is not a simple one. However, I will conclude by
offering a few suggestions for steps that can be taken in the direction that the concep-
tual design perspective implies.
A first, monumentally difficult step in this literature would be agreement on termi-
nology. If the study of concepts is to achieve consilience with work in neuroscience,
genetics, development, and comparative biology, then it is of crucial importance that
we (eventually) agree on what the technical concept CONCEPT is meant to pick out in
the mind. Machery (2009), seeking to formalize such a consensus, despairs at the
possibility of ever doing so, and to some extent I share his pessimism. However, I
suggest that the hierarchical view of mental structure that I have presented here and
elsewhere (Barrett 2012, 2015) provides a way of naturalizing concepts without essen-
tializing them. What I mean by this is that like other biological concepts that pick
out putative natural kinds—or at least kinds that biologists make use of in their daily
work, without referential gridlock—a biological theory of concepts can be situated
within a view of biological traits as evolving via descent with modification, as consti-
tuted of smaller parts or features embedded within larger networks of parts or features,
and to some degree as fuzzy in their boundaries. On this view we can treat conceptual
systems as just that—systems—in the biological and physical sense. As part of this
view, we can use the language of types and tokens, without privileging any level of the
hierarchy as the conceptual level: my momentary tokening of an individual doughnut
representation could be spoken of as a concept, as well as my general concept DOUGHNUT
and the higher-level types or categories such as FOOD and OBJECT in which DOUGHNUT is
A second, crucial step is the empirical documentation of conceptual diversity
and richness across individual humans, across human societies, and across species—
necessary for a proper evolutionary, comparative perspective on the evolution of
concepts. To a large degree, of course, this is already under way: it’s what much of
psychology, anthropology, neuroscience, and behavioral biology already do. What is
lacking, however, is a common framework in which to fit these data together, bringing
them to bear on testable theories of conceptual design. What is needed is some way
of standardizing data from different cultures, stages in development, and species. In
neuroscience and genetics, promising steps are being taken toward standardized data
formats through the construction of formal ontologies: standardized classifications of
brain structures (Price and Friston 2005), genes (Ashburner et al. 2000), and the links
between them (Bilder, Sabb, Cannon, et al. 2009). In the case of cognitive ontologies
used in brain mapping, these are explicitly about phenomics: establishing standardized
The Evolution of Conceptual Design 177

classifications of cognitive phenotypes in developed brains, and linking behavioral,

brain mapping, and genetic data (Bilder, Sabb, Parker, et al. 2009). Some such form
of standardization of conceptual knowledge, across individuals, species, and develop-
mental stages, would greatly assist the construction of biologically accurate reaction
norms of cognitive development. In addition, a move away from null hypothesis test-
ing and toward model comparison—testing theories against each other, rather than
against often implausible nulls—will greatly increase our ability to compare the predic-
tive power of alternative theories, statistically controlling for their relative complexity
(Burnham and Anderson 2002).
In order to do this, of course, we’ll need actual theories: computationally specified
models of the developmental reaction norms that produce conceptual phenotypes.
Perhaps the most fully fleshed out of such theories to date are Bayesian models of
conceptual development, because these specify, in principle, the complete develop-
mental trajectory from infancy to adulthood given some set of environmental inputs
(Kemp and Tenenbaum 2008 Tenenbaum, Griffiths, and Kemp 2006; Tenenbaum et al.
2011). However, even these existing models are almost certainly too simple by orders
of magnitude, to the extent that they leave out factors like content specificity, brain
architecture, and interactions between the brain’s specialized subsystems during devel-
opment. Models that actually specify a developmental curve of any kind, though, are a
step in the right direction. Eventually, these must be situated in a phylogenetic context,
to account for descent with modification of human conceptual structures, as well
as in the context of developmental gene expression and neural development. At
present, there is an enormous gap between the psychological literature on concepts
and conceptual development, and the biologically based literatures on brain organiza-
tion and development—though this gap is beginning to close (e.g., Amodio and Frith
2006; Saxe, Carey, and Kanwisher 2004).
Finally, in keeping with the above, our theories of human conceptual development
will need to merge with contemporary evo-devo views of how biological structures
evolve and develop (Carroll, Grenier, and Weatherbee 2005). While human conceptual
development is often treated as a sui generis phenomenon, frequently starting from
philosophical considerations, clearly whatever conceptual capacities we have must
have evolved through descent with modification from ancestral structures. Descent
with modification leads to, among other things, nested hierarchies of traits: the diverse
and functionally differentiated structures of human brains, for example, developed
through evolutionary differentiation from a smaller number of ancestral structures.
This means that structures such as the cortex are hierarchically organized, with some
features shared in common and others functionally divergent (Allman 2000; Barrett,
2012; Krubitzer and Huffman 2000). Because conceptual structure is produced by brain
structure, the same is likely true of our conceptual repertoires and the developmental
machinery that builds them: the richness of modern-day human conceptual repertoires
178 H. Clark Barrett

likely evolved through descent with modification from simpler, ancestral conceptual
repertoires. Thus, the conceptual palette of our lineage likely bifurcated and prolifer-
ated over evolutionary time, both prior to and after our split from chimpanzees, to
become more rich and diverse. But, just like our limbs, livers, and frontal lobes, every
concept we possess is likely to retain traces of the ancestral pathway of descent with
modification it followed. With sufficient care, we can use these facts about the histori-
cal nature of the evolutionary process to build more fine-grained and biologically accu-
rate models of human conceptual development.
As an example of the insight such a hierarchical view might add, consider debates
about the evolution of theory of mind, or mindreading, and the array of mental-state
concepts, such as BELIEF, DESIRE, and KNOWLEDGE, that are thought to underpin this ability
(Nichols and Stich 2003). On some accounts, this is an entirely unique human phe-
nomenon that arose de novo in the human lineage, with no evolutionary continuity
with other taxa (Penn and Povinelli 2007). However, this seems unlikely: although
there may be many unique aspects of human mindreading, it seems likely that human
mindreading, like all aspects of human cognition, evolved through descent with modi-
fication from ancestral cognitive abilities and is therefore likely to have homologs in
other species—including at least some homologous concepts. Comparative research
suggests that this is the case, with other species able to interpret cues of communica-
tive intent, such as mating and aggression, and to track indices of mental states such
as knowledge. This suggests that there is likely to be some homology in the underly-
ing mechanisms, and in the representations—including concepts—they employ (Call
and Tomasello 2008; Hare et al. 2002; Clayton, Dally, and Emery 2007). On the view
presented here, this does not require that a chimpanzee’s concept of knowledge or
desire need be identical to ours; but the hypothesis that they are evolutionarily related,
or homologous, is likely to shed light on the evolution and development of the brain
mechanisms underlying mindreading in our species and theirs.
What is perhaps surprising, given the long history of research and theorizing about
the human conceptual repertoire, is how much remains unknown about it. Arguably
this results partly from insisting that the story of human conceptual development be a
relatively simple one, accounted for by a small number of conceptual building blocks
interacting with a complex world. Here I have proposed a new approach, the con-
ceptual design approach, which starts with the functional roles that concepts play in
our developed conceptual repertoires and asks how the developmental resources that
build these repertoires are shaped by the evolutionary process. Although this is a novel
approach and therefore untested, it has the potential to offer new ways of merging
the study of concepts with work in neuroscience, brain evolution, and development.
By allowing for the possibility that the mind’s conceptual repertoire may be built by a
diverse family of mechanisms with many possible designs and evolutionary histories,
The Evolution of Conceptual Design 179

we may discover that the mind’s complexity is richer than previously thought—and
that it is rich by design.


Allman, J. M. 2000. Evolving Brains. New York: Scientific American Library.

Amodio, D. M., and C. D. Frith. 2006. Meeting of minds: The medial frontal cortex and social
cognition. Nature Reviews: Neuroscience 7 (4): 268–277.

Anderson, M. L. (2010). Neural reuse: A fundamental organizational principle of the brain.

Behavioral and Brain Sciences, 33, 245–266.

Ariew, A. 1996. Innateness and canalization. Philosophy of Science 63:19–27.

Ashburner, M., C. A. Ball, J. A. Blake, D. Botstein, H. Butler,J. M. Cherry, et al. 2000. Gene ontol-
ogy: Tool for the unification of biology. Nature Genetics 25 (1): 25–29.

Atkinson, A. P., and M. Wheeler. 2004. The grain of domains: The evolutionary-psychological
case against domain-general cognition. Mind & Language 19 (2): 147–176.

Baillargeon, R. 1994. How do infants learn about the physical world? Current Directions in Psycho-
logical Science 3 (5): 133–140.

Barrett, H. C. 2005a. Enzymatic computation and cognitive modularity. Mind & Language 20 (3):

Barrett, H. C. 2005b. Adaptations to predators and prey. In The Handbook of Evolutionary Psychol-
ogy, ed. D. M. Buss, 200–223. New York: Wiley.

Barrett, H. C. 2006. Modularity and design reincarnation. In The Innate Mind: Culture and Cogni-
tion, ed. P. Carruthers, S. Laurence, and S. Stich, 199–217. New York: Oxford University Press.

Barrett, H. C. 2007. Development as the target of evolution: A computational approach to

developmental systems. In The Evolution of Mind: Fundamental Questions and Controversies, ed.
S. Gangestad and J. Simpson, 186–192. New York: Guilford.

Barrett, H. C. 2009. Where there is an adaptation, there is a domain: The form-function fit in
information processing. In Foundations in Evolutionary Cognitive Neuroscience, ed. S. M. Platek and
T. K. Shackelford, 97–116. Cambridge: Cambridge University Press.

Barrett, H. C. 2012. A hierarchical model of the evolution of human brain specializations. Proceed-
ings of the National Academy of Sciences of the United States of America 109:10733–10740.

Barrett, H.C. 2015. The Shape of Thought: How Mental Adaptations Evolve. New York: Oxford Uni-
versity Press.

Barrett, H. C., and J. Broesch. 2012. Prepared social learning about dangerous animals in children.
Evolution and Human Behavior 33 (5): 499–508.
180 H. Clark Barrett

Barsalou, L. W. 1999. Perceptual symbol systems. Behavioral and Brain Sciences 22 (4): 577–660.

Barsalou, L. W. 2005. Continuity of the conceptual system across species. Trends in Cognitive
Sciences 9 (7): 309–311.

Bilder, R. M., F. W. Sabb, T. D. Cannon, E. D. London, J. D. Jentsch, D. S. Parker, et al. 2009.

Phenomics: The systematic study of phenotypes on a genome-wide scale. Neuroscience 164 (1):

Bilder, R. M., F. W. Sabb, D. S. Parker, D. Kalar, W. W. Chu, J. Fox, et al. 2009. Cognitive ontolo-
gies for neuropsychiatric phenomics research. Cognitive Neuropsychiatry 14 (4–5): 419–450.

Blaisdell, A. P., K. Sawa, K. J. Leising, and M. R. Waldmann. 2006. Causal reasoning in rats. Science
311 (5763): 1020–1022.

Boroditsky, L. 2001. Does language shape thought? Mandarin and English speakers’ conceptions
of time. Cognitive Psychology 43 (1): 1–22.

Boyer, P. 2002. Religion Explained. New York: Basic Books.

Burnham, K. P., and D. R. Anderson. 2002. Model Selection and Multi-Model Inference: A Practical
Information-Theoretic Approach. New York: Springer.

Call, J., and M. Tomasello. 2008. Does the chimpanzee have a theory of mind? 30 years later.
Trends in Cognitive Sciences 12 (5): 187–192.

Caramazza, A., and B. Z. Mahon. 2003. The organization of conceptual knowledge: The evidence
from category-specific semantic deficits. Trends in Cognitive Sciences 7 (8): 354–361.

Carey, S. 2009. The Origin of Concepts. New York: Oxford University Press.

Carroll, S. B., J. K. Grenier, and S. D. Weatherbee. 2005. From DNA to Diversity: Molecular Genetics
and the Evolution of Animal Design. 2nd ed. Oxford: Blackwell.

Changizi, M. A., Q. Zhang, H. Ye, and S. Shimojo. 2006. The structures of letters and symbols
throughout human history are selected to match those found in objects in natural scenes. Ameri-
can Naturalist 167 (5): E117–E139.

Charnov, E. L. 1976. Optimal foraging, the marginal value theorem. Theoretical Population Biology
9 (2): 129–136.

Cheney, D. L., and R. M. Seyfarth. 2007. Baboon Metaphysics: The Evolution of a Social Mind.
Chicago: University of Chicago Press.

Chiappe, D., and K. MacDonald. 2005. The evolution of domain-general mechanisms in intelli-
gence and learning. Journal of General Psychology 132 (1): 5–40.

Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.

Clayton, N. S., J. M. Dally, and N. J. Emery. 2007. Social cognition by food-caching corvids: The
western scrub-jay as a natural psychologist. Philosophical Transactions of the Royal Society of London:
Series B, Biological Sciences 362 (1480): 507–522.
The Evolution of Conceptual Design 181

Cohen, L., and S. Dehaene. 2004. Specialization within the ventral stream: The case for the visual
word form area. NeuroImage 22 (1): 466–476.

Dehaene, S. 2009. Reading in the Brain: The Science and Evolution of a Human Invention. New York:

Fodor, J. A. 1998. Concepts: Where Cognitive Science Went Wrong. New York: Oxford University

Gallistel, C. R., and A. P. King. 2009. Memory and the Computational Brain: Why Cognitive Science
Will Transform Neuroscience. New York: Wiley-Blackwell.

Gopnik, A., and A. Meltzoff. 1997. Words, Thoughts, and Theories. Cambridge, MA: MIT Press.

Gottlieb, G. 2006. Probabilistic epigenesis. Developmental Science 10 (1): 1–11.

Griffiths, P. E., and E. Machery. 2008. Innateness, canalization, and ‘biologicizing the mind’.
Philosophical Psychology, 21(3): 397–414.

Griffiths, T. L., N. Chater, C. Kemp, A. Perfors, and J. B. Tenenbaum. 2010. Probabilistic models
of cognition: Exploring representations and inductive biases. Trends in Cognitive Sciences 14 (8):

Hare, B., M. Brown, C. Williamson, and M. Tomasello. 2002. The domestication of social cogni-
tion in dogs. Science 298 (5598): 1634–1636.

Heibeck, T. H., and E. M. Markman. 1987. Word learning in children: An examination of fast
mapping. Child Development 58 (4): 1021–1034.

Hughes, A. L. 1994. The evolution of functionally novel proteins after gene duplication. Proceed-
ings of the Royal Society of London: Series B, Biological Sciences 256 (1346): 119–124.

Jackendoff, R. 1996. The architecture of the linguistic-spatial interface. In Language and Space, ed.
P. Bloom, M. A. Peterson, L. Nadel, and M. F. Garrett, 1–30. Cambridge, MA: MIT Press.

Johnson-Frey, S. H. 2004. The neural bases of complex tool use in humans. Trends in Cognitive
Sciences 8 (2): 71–78.

Kanwisher, N. 2010. Functional specificity in the human brain: A window into the functional
architecture of the mind. Proceedings of the National Academy of Sciences of the United States of
America 107 (25): 11163–11170.

Keil, F. C. 1992. Concepts, Kinds, and Cognitive Development. Cambridge, MA: MIT Press.

Kemler Nelson, D. G., R. Russell, N. Duke, and K. Jones. 2003. Two-year-olds will name artifacts
by their functions. Child Development 71 (5): 1271–1288.

Kemp, C., and J. B. Tenenbaum. 2008. The discovery of structural form. Proceedings of the National
Academy of Sciences of the United States of America 105 (31): 10687–10692.

Krubitzer, L., and K. J. Huffman. 2000. A realization of the neocortex in mammals: Genetic and
epigenetic contributions to the phenotype. Brain, Behavior and Evolution 55 (6): 322–335.
182 H. Clark Barrett

Lakoff, G., and M. Johnson. 1980. Metaphors We Live By. Chicago: University of Chicago Press.

Laurence, S., and E. Margolis. 1999. Concepts and cognitive science. In Concepts: Core Readings,
ed. E. Margolis and S. Laurence, 3–81. Cambridge, MA: MIT Press.

Leslie, A. M. 1994. Pretending and believing: Issues in the theory of ToMM. Cognition 50 (1):

Liebenberg, L. 1990. The Art of Tracking: The Origin of Science. Claremont, South Africa: D. Philip.

Machery, E. 2009. Doing Without Concepts. New York: Oxford University Press.

Margolis, E., and S. Laurence. 2007. The ontology of concepts—abstract objects or mental repre-
sentations? Noûs 41 (4): 561–593.

Marr, D. 1982. Vision: A Computational Investigation into the Human Representation and Processing Of
Visual Information. New York: Henry Holt.

Murphy, G. L. 2002. The Big Book of Concepts. Cambridge, MA: MIT Press.

Nichols, S., and S. Stich. 2003. Mindreading. New York: Oxford University Press.

Penn, D. C., and D. J. Povinelli. 2007. On the lack of evidence that non-human animals possess
anything remotely resembling a “theory of mind.” Philosophical Transactions of the Royal Society of
London: Series B, Biological Sciences 362 (1480): 731–744.

Pinker, S. 1989. Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, MA:
MIT Press.

Prasada, S., K. Ferenz, and T. Haskell. 2002. Conceiving of entities as objects and as stuff.
Cognition 83 (2): 141–165.

Price, C. J., and K. J. Friston. 2005. Functional ontologies for cognition: The systematic definition
of structure and function. Cognitive Neuropsychology 22 (3–4): 262–275.

Pylyshyn, Z. W. 1984. Computation and Cognition. Cambridge, MA: MIT Press.

Richerson, P., and R. Boyd. 2005. Not by Genes Alone: How Culture Transformed Human Evolution.
Chicago: University of Chicago Press.

Rosch, E. 1999. Principles of categorization. In Concepts: Core Readings, ed. E. Margolis and
S. Laurence, 189–206. Cambridge, MA: MIT Press.

Samuels, R. 2002. Nativism in cognitive science. Mind & Language 17 (3): 233–265.

Saxe, R., S. Carey, and N. Kanwisher. 2004. Understanding other minds: Linking developmental
psychology and functional neuroimaging. Annual Review of Psychology 55:87–124.

Schlichting, C. D., and M. Pigliucci. 1998. Phenotypic Evolution: A Reaction Norm Perspective. New
York: Sinauer Associates.

Sherry, D. F., and D. L. Schacter. 1987. The evolution of multiple memory systems. Psychological
Review 94 (4): 439–454.
The Evolution of Conceptual Design 183

Smith, E. A. 1983. Anthropological applications of optimal foraging theory: A critical review.

Current Anthropology 24 (5): 625–651.

Smith, E. E., and D. L. Medin. 1981. Categories and Concepts. Cambridge, MA: Harvard University

Spelke, E., and K. Kinzler. 2009. Innateness, learning, and rationality. Child Development Perspec-
tives 3 (2): 96–98.

Sperber, D. 1996. Explaining Culture: A Naturalistic Approach. Oxford: Blackwell.

Stich, S. P. 1983. From Folk Psychology to Cognitive Science: The Case against Belief. Cambridge, MA:
MIT Press.

Tenenbaum, J. B., T. L. Griffiths, and C. Kemp. 2006. Theory-based Bayesian models of inductive
learning and reasoning. Trends in Cognitive Sciences 10 (7): 309–318.

Tenenbaum, J. B., C. Kemp, T. L. Griffiths, and N. D. Goodman. 2011. How to grow a mind:
Statistics, structure, and abstraction. Science 331 (6022): 1279–1285.

Tooby, J., and L. Cosmides. 1990. The past explains the present: Emotional adaptations and the
structure of ancestral environments. Ethology and Sociobiology 11 (4): 375–424.
7 How Natural Selection Shapes Conceptual Structure: Human
Intuitions and Concepts of Ownership

Pascal Boyer

7.1 Mapping Concepts from World, Language, … and Evolution

How do we map the inventory of human concepts? Here I propose that a precise
description of selective pressures on species-specific cognitive systems is the best source
of empirical hypotheses about conceptual repertoires, and I illustrate this in the case of
ownership concepts. The hypotheses presented here stem from a naturalistic agenda, in
which concepts like other mental phenomena are construed as functional properties of
cognitive systems (Jackendoff 1983; Margolis and Laurence 2007; Millikan 1998). The
example of ownership illustrates how a highly specific selective context can predict
and explain equally specific aspects of human concepts. This account also suggests
more general though tentative lessons, to do with what general computational prop-
erties, if any, should be expected from concepts; whether categorization is crucial to
concept structure; and what role concepts play in linguistic reference.
The evolutionary perspective stands in contrast to two other possible ways of
proceeding. One is a form of realism that is often implicit in psychological research
on conceptual knowledge, and is often combined with the discipline’s unreflective
empiricism (Carey 2009, 27ff.). This perspective assumes that we can infer a cognitive
system’s conceptual repertoire from our knowledge of the world that system is embed-
ded in. After all, the argument goes, to survive, complex organisms must entertain
beliefs that at least roughly match the way the world is. This makes it likely that many
organisms’ systems of concepts “carve nature at the joints,” to use the common phrase.
So the fact that there are actually different kinds of things in the world makes it likely
that concepts reflect those distinct categories. For instance, plants and animals are
substantially different, so any animal with a complex cognitive system would probably
have general concepts for plants and animals.
However, nature does not actually have “joints” that would all be equally rele-
vant to organisms from different species. For instance, it is unlikely that dogs have a
fully general concept ANIMAL, because they do not interact with all animals on the
basis of a single set of expectations and motivations. One would expect that dogs
186 Pascal Boyer

probably have at least the following distinct concepts: HUMAN, DOG, PREY, because each
of these kinds of agents activates a distinct suite of cognitive systems. Being domesti-
cated by humans, dogs interact with them based on cognitive resources that are not
used for dealing with other dogs or with prey. But other systems in the dog’s mind, for
example, mating systems, are uniquely activated by other dogs and not by any other
living thing.
A second standard approach is to try to infer a conceptual repertoire from natural
language, taking concepts to be stable information structures that correspond to lexical
units. This linguistic perspective has dominated general research on concepts, so I will
not dwell on its essential tenets and many achievements. I should just point to some
of its limitations. First, as Sperber and Wilson point out, though it is clear that human
minds contain as many concepts as lexical items (barring genuine synonyms), it is quite
plausible that they have vastly more concepts (Sperber and Wilson 1998, 275), even if
we limit ourselves to propositionally encoded information. Second, humans have a
large number of ineffable concepts, like modality-encoded information, that organize
information in a stable way yet cannot be expressed linguistically (Barsalou et al. 2003).
Third, the linguistic turn suggests a radical hiatus between human and other-animal
cognition, which ignores behavioral homologies and plausible phylogenetic continu-
ity between species.
Here I will outline and illustrate a third approach in mapping the conceptual land-
scape, one that starts from selective pressures over evolutionary history, and formulates
hypotheses about optimal and feasible cognitive structures that would respond to these
pressures in fitness-enhancing ways. Considering selective pressures on organisms pro-
vides a lot of information about what their concepts may be like and therefore many
rich hypotheses about cognitive architecture that we can test (Tooby, Cosmides, and
Barrett 2005). Indeed, one point of this article is to suggest that such evolutionary task
analysis provides much more information than we would first imagine, and therefore
should be used as an important source for an evolutionary account of concepts.

7.2 Ownership as a Conceptual Domain

Ownership is central to human social interaction. Among the culturally universal

aspects of ownership are the fact that (a) all known human languages can express,
through grammatical or lexical means or both, the special connection between spe-
cific agents and particular things, either material (natural objects, artifacts, territories)
or abstract (songs, stories, knowledge); (b) there is a principled distinction between
mere possession and ownership; (c) ownership is associated with specific emotions
and motivations; (d) the acquisition of ownership notions and norms occurs very early
and effortlessly, along highly similar developmental lines (Brown 1991; Heine 1997).
By contrast, norms of ownership and property rights differ from one place or time to
How Natural Selection Shapes Conceptual Structure 187

another in terms both of scope (who can own things and what things can be owned)
and of implications (what one may do with specific types of property) (Hann 1998).
So how do we approach the conceptual underpinnings of these linguistic, mental,
and social phenomena? The first method to concepts mentioned above, from actual
kinds to constraints on concepts, is obviously not available here. There is no physi-
cal fact of the matter that corresponds to ownership. Whether an agent is the owner
of a thing or not is the consequence, not of physical properties of agent and thing,
but of shared mental representations between the agent and third parties. The second
method, from lexicon and syntax to concepts, provides suggestive but ambiguous
information. Ownership is part of a much broader domain of possession relations that
includes kin, body parts, and other non-accidental appendages, as it were, of an agent.
Possession can be conveyed by attributive means (e.g., “her blue eyes,” “Jane’s blue
eyes”) or predicative means (“she has blue eyes”) (Heine 1997, 25ff.; Rudmin 1994).
The linguistic evidence suggests that possession constructions, and a fortiori specific
ownership constructions, are derived from nonpossession semantic structures (Heine
1997, 224), for example, from spatial schemas. For instance, OWN(AGENT, THING) can be
seen as a special version of a BE NEXT TO(AGENT, THING) concept, and GIVE(AGENT1, AGENT2,
THING) is analogous to MOVE(AGENT1, AGENT2, THING), and so on (Jackendoff 1983). These

models also illustrate why semantic structures are only of limited value if we want to
elucidate conceptual organization. Mappings, such as from spatial relations to owner-
ship, occur in natural languages because they are intuitively appropriate—which is the
case because of specific underlying assumptions about ownership, that are precisely
what we want to describe.
Surprisingly, there was until recently very little systematic research on the psychol-
ogy of ownership—on the intuitions and explicit thoughts engaged when people make
claims about who owns what or how property can be transferred (Friedman 2010).
Most of the relevant recent evidence comes from developmental psychology. Children
have clear and specific intuitions about ownership and, of course, very strong motiva-
tions associated with those intuitions. Claims that a particular object is mine appear
very early in young children’s verbal communication, and children engage in frequent
disputes over ownership. Although ownership is invisible, children readily infer it
from verbal information to the effect that a particular object is theirs or belongs to
another person, and they adjust their behavior accordingly, from thirty-six months
(Eisenberg-Berg, Haake, and Bartlett 1981; Ross and Conant 1992), or even from twenty-
four months (Blake, Ganea, and Harris 2012). What conceptual structures underpin
this competence?
Children’s explicit statements about ownership are not altogether consistent, and
are sometimes downright odd. Children, for instance, agree that some objects can be
owned, like artifacts or natural objects extracted from the environment, but they are
less certain about abstract ideas. More surprising, four year-olds state that sleeping
188 Pascal Boyer

people cannot “own” objects (Noles and Keil 2011). This may suggest that whatever
ownership concepts have developed at that point, they are only weakly connected to
the lexical items own or belong.
In contrast with these generic notions, even very young children have definite
intuitions about the specific events that create or transfer ownership. In the absence
of verbal information, children rely on a first possessor heuristic—whoever used or
handled the object first is presumed to be the owner (Blake and Harris 2009; Friedman
and Neary 2008). First possessors usually win conflicts over toys at age three, but not
at age one (Bakeman and Brownlee 1982). Children’s intuitions also imply that effort
invested in some object results in ownership. Preschoolers, for instance, judge that
an agent A’s block of clay belongs to B if B sculpted it, but not if B merely handled it
without changing its shape (Kanngiesser, Gjersoe, and Hood 2010). These and other
studies show that from thirty-six months or earlier, children are familiar with the
exclusion aspect of ownership. They also have stable intuitions regarding transfers of
ownership, if these are made sufficiently explicit (Friedman and Neary 2008; Kim and
Kalish 2009).
We find a similar contrast between explicit conceptions and intuitions in adults.
In response to questionnaires, for instance, adults readily assert that persons cannot
be owned (before being reminded of the history of slavery). Like children, they state
that specific information can be owned, but not generic knowledge, although they
cannot elaborate on the distinction (Noles and Keil 2011). Indeed, questions about
what kinds of objects can or cannot be owned always trigger vague or inconsistent
answers—because it is not clear in what sense ownership could be restricted to any
particular ontological domain (Friedman 2010). But adults like children have reliable
intuitions about use and possession, for example, they use a first possessor heuristic to
determine ownership (Friedman et al. 2011), as do many legal systems the world over—
as the saying goes, “possession is 90 percent of the law.” This is not absolute, though.
Friedman and colleagues demonstrated that intuitions about use and possession are
largely dependent on cues concerning the history of the object, that is, who made it
and how, or who extracted it from natural nonowned resources, what transfers took
place, how much effort was involved, and so forth (Friedman et al. 2011). Also, adults’
intuitions are sensitive to cues that make one agent responsible for the object being
around, or being available, or having a utility, and responsibility here depends on an
appreciation of the agent’s intentions when he or she displaced or modified the object
(Palamar, Le, and Friedman 2012).
The psychological evidence, however, does not by itself explain why ownership is
construed in this way, why people’s intuitions are only imperfectly tracked by their
explicit concepts, and why facts about an object’s history modulate people’s intuitions.
To address these questions, one must consider the adaptive functions of these concep-
tual structures.
How Natural Selection Shapes Conceptual Structure 189

7.3 Possession and Ownership (I): Intuitions and Motivations

An evolutionary approach implies that cognitive systems constitute evolved responses

to recurrent challenges facing organisms in their evolutionary environments. In the
case at hand, I propose (along with others; see, e.g., DeScioli and Wilson 2011) that
ownership as a conceptual domain is part of our responses to the fundamental chal-
lenge of reaching a measure of coordination that optimizes the extraction of resources.

7.3.1 Ownership Intuitions as an Evolved Coordination Strategy

Humans like other organisms from highly social species can extract resources from
their environments better if they avoid a Hobbesian war of all against all, in which the
acquisition and use of resources are extremely costly, as every item must be extracted
from the environment under the threat of fights with, and potential theft by, con-
specifics. That may be why this alleged “state of nature” is not actually very natural.
Organisms from many species manage to avoid its pitfalls through coordination, that
is, a broad class of strategies in which most organisms abide by some norm of resource
extraction—for instance, that the first agent to occupy a territory will keep it, while
others will try to find another place (Maynard Smith 1982). Having such norms greatly
reduces the costs of resource extraction, thereby increasing each organism’s fitness
(Johnsson, Nöbbelin, and Bohlin 1999). (Note that adopting coordination strategies
in this perspective is a matter of genetic fitness, not group survival—it does not matter
if the strategy benefits groups or not). The pressure for coordination norms is espe-
cially acute in humans, who are more dependent than any other species on interaction
with conspecifics to acquire resources from their environment, and have evolved the
required capacities for sophisticated coordination in such endeavors as warfare, hunt-
ing, co-parenting, and many others (Dubreuil 2010; Gat 2006; Hrdy 2009).
The term resource should be taken in the broadest sense, as any part of the envi-
ronment, interaction with which can increase the organism’s fitness. This includes
food, shelter, mates, and potential allies. Now many of the resources extracted are rival
goods, such that one agent’s enjoyment of the resource diminishes other agents’ poten-
tial use. Many of these rival goods are also potentially excludable, such that an agent
can to some extent bar another agent from access to a particular resource.
Trade and gift giving impose additional evolutionary pressures on ownership cogni-
tion. Trade appeared early in human evolution—indeed it may be one of the evolution-
ary innovations that mark the advent of modern human societies, as well as providing
a significant boost to creativity and human innovation (Ridley 2010). As for gift giving,
it is a universal feature of human societies. Social relations are created or cemented by
gifts. Because of the selective pressures for appropriate understanding of trade and gift
giving, human intuitions of ownership should not be immutably linked to an object’s
history. There should be some possibility for A’s ownership of x to be transferred to
190 Pascal Boyer

some other agent. Note, however, that this creates a series of coordination problems.
First, if A transfers x to B, the extent to which A is not the owner of x should be con-
strued in the same way by both parties. Second, A and B should have some means of
knowing at precisely what point B owns x. Finally, third parties should also receive the
information such that they adjust their behaviors accordingly.
The proposal here is that intuitions and motivations concerning who uses what
resources constitute such a coordination tool for humans—and evolved from less
efficient coordination strategies in the course of human evolution. In this perspective,
the complex of intuitions and motivations generally called ownership are the outcome
of largely tacit computations concerning the relative costs and benefits of using, guard-
ing, or poaching resources, as well as collaborating with others in these diverse courses
of action.

7.3.2 The Contents of Possession-Related Intuitions

Ownership, like other aspects of social interaction, is a domain for which normal
human beings have specific, largely automatic intuitions, that is, mental represen-
tations of a particular state of affairs, or motivations to act toward particular states,
without a clear or accessible representation of the processes that lead to these intu-
itions. In the same way as depth perception requires largely unconscious principles to
organize visual information, ownership-relevant information in the environment can
be processed outside conscious access to yield specific intuitions. Imagine the following
scene: Robinson picks up some shells and seaweed on the beach and assembles them
as a bracelet that he puts down. Friday handles the bracelet. The scene triggers in most
third parties some intuitive representations to the effect that “this is Robinson’s brace-
let,” “this is not Friday’s bracelet.”
The intuitions we consider here are all about some specific agent using some spe-
cific resource. Faced with some resource, the organism A has a choice among courses
of action that we can label TAKE, NOT-TAKE, DEFEND, or RELINQUISH, each with its associated
costs and benefits. Faced with two agents A and B, one of whom is using a resource,
there is a choice of actions for C, dependent on the interaction between A and B. If
A takes from B, there is a choice among DO NOTHING, HELP TAKE, HELP GUARD, and so on.
(Although this seems to invite infinite regress, there may be no selective pressure for
any such coordination beyond triadic interaction, as will be discussed presently.) These
are the strategic choices. What is the content of the intuitions about these choices?
Our hypothesis here is that, faced with such situations, organisms are equipped with
a domain-specific system that takes as input certain cues, as well as some background
information, to produce definite intuitions and motivations toward TAKE(x), GUARD(x),
HELP TAKE(x), HELP GUARD(x), and so forth.

These intuitions and motivations are principled. The domain-specific system

involved is such that in response to specific patterns of situations, it produces predict-
able intuitions and motivations that guide the organism’s behavior. The organization
How Natural Selection Shapes Conceptual Structure 191

of these intuitions instantiates strategic conventions, that is, evolutionarily stable strat-
egies. They help organisms save energy by avoiding fights for resources and expend
that energy on seeking alternative resources.
Note that the intuitions and motivations of different agents are, of course, not
always matched. That is, coordination may break down and encounters lead to a
number of fights. This may happen because the cues processed by two organisms are
different (e.g., an organism actually is the first occupant of a territory, but an intruder
failed to perceive that). This may also happen because the potential rewards of fights in
some cases outstrip the motivation for strategic withdrawal. That would be the case, for
example, if your friend knows that your lottery ticket is a winning one.

7.3.3 Computational Properties of Ownership Intuitions

Note the following important points about the mental representations involved:
This is all intuitive. All the mental content I have described so far comes in the form
of intuitions, that is, mental content that is not the product of deliberate ratiocination;
nor is it necessarily explicit. What makes it intuitive is that some representations and
motivations, for example, “Don’t take berries from this bush!” or “Attack anyone who
tries to use this tool!” are entertained without any representation of the processes that
led to them. This should be emphasized because intuitions in this sense are contrasted
below with reflective mental content.
Intuitions and motivations are outputs of the same systems. The mental representations
described here come as a package, in which a description of a state of affairs and a drive
to make a certain state of affairs real are intertwined. A particular situation triggers a
mental representation that is both description (e.g., Individual A is using object x) and
motivation (e.g., take x away from A!).
Intuitions are domain specific. So far, we have considered common features in all such
intuition systems, regardless of their domain of operation. That is why this description
is bound to remain extremely abstract. To go further in the description of these cogni-
tive processes, we may need to consider specific domains of such resources, such as
mating, foraging, hunting, or tool making. Different domains of behavior, in a highly
social species, result in different challenges in terms of access to resources and exclu-
sion of other agents, such that it is unlikely that an optimal system would just realize
one single set of principles for them all.
Computations explain norms (not the other way around). One might want to object
that the model proposed does not so far capture the normative aspects of ownership.
In describing ownership psychology, most attempts at either evolutionary or historical
construction start from a rich notion of norms applied to possession and use. Indeed,
this may seem the most natural starting point. After all, a concept of ownership should
make sense of cases in which someone is but should not be in possession of an object,
and even young children grasp the point. One goal of the present research program,
however, is precisely to elucidate the various computational processes leading to, for
192 Pascal Boyer

example, a strong motivation to have the thief restitute stolen goods to their previous
possessors, or to a specific feeling of frustration when people take away what we made
without compensation. These (and many other) specific emotions and intentions
underpin what we call ownership norms, but calling them that does not in any sense
illuminate the processes involved. The deontic intuition is precisely what we should
explain. Instead of assuming, for example, that representations about how a resource
was extracted feeds into an ownership norm, I choose here to consider how cues about
history (e.g., an agent A extracted object x from nature, or agent A invested work into
x, etc.) directly trigger motivations and intuitions to the effect that A keeps x, that one
is not motivated to try to appropriate x, that others may want to defend A against such
exploitation, and so on, which are motivating, predictable, and shared—the features
of what we call norms.

7.4 Coordination Proposal (II): Explicit Beliefs and Norms

Beyond intuitive representations, people in some circumstances also entertain reflec-

tive representations of ownership-relevant information. These are explicit, generally
verbalized series of representations that explicate, amplify, concatenate, or comment
on intuitions. To reprise our example, these would be explicit representations like “It is
R’s bracelet because R made it,” “It is not F’s bracelet as R did not give it to F,” or “It is
not Y’s necklace because X only lent it to Y.” Note that one does not require the explicit
reflections to entertain the intuitions. Young children, for instance, will share some of
the intuitions mentioned above without being able to articulate their rationale. This
distinction between intuitions and reflections is similar to that between intuitive and
reflective beliefs (Sperber 1997), aliefs and beliefs (Gendler 2008), or architectural beliefs
and metarepresentations (Cosmides and Tooby 2000).

7.4.1 Why Bother with Explicit Representations?

Why have explicit, universally lexicalized concepts as a complement to (and loosely
associated with) specific intuitions? The requirements for extensive coordination
among humans make this development, if not inevitable, at least very likely in most
human groups. This is because coordination requires compatible motivations, but also
shared information. It requires that, once agent X has extracted resource R, most third
parties that were not present during resource acquisition adopt the same noninterfer-
ence strategy as the parties present. In other words, it requires some signal that sums
up what can be inferred from the history of acquisition, and it triggers relevant moti-
vations in other agents. This requirement is, of course, not unique to humans. For
instance, many animals need to signal that they are “owners” of a particular territory
rather than passersby, which they usually convey by specific signals. Humans accom-
plish that in the species-specific manner that matches their much greater requirements
How Natural Selection Shapes Conceptual Structure 193

for coordination, through verbal communication of the connection between agent and
thing possessed.
Explicit beliefs about ownership do not easily track the subtle contextual cues that
modify our intuitions. Consider the familiar case of jokes. Most people would assume
that a joke is not something you can “own.” So telling a joke you heard is not taken as
an instance of plagiarizing, of stealing anyone’s property. But the intuition can change.
On our way to a fancy dinner, you tell me a great joke that you hope will dazzle the
crowd. But as we sit down at the table, I tell that precise joke. Now it seems that in a
sense I did take something that was “yours.” Even clearer, of course, is the case of a
comedian who uses a colleague’s material—that is definitely stealing. Why do our intu-
itions change with the context, given that the actions are similar? An easy (and wrong)
answer is that the actions are not similar, because in the case of the comedian (vs.
ordinary conversations), a joke is “intellectual property.” But that is question begging—
we readily assign that label to comedian’s jokes precisely because we have intuitions,
for example, that they have a special entitlement to getting benefits from their mate-
rial, that we should approve them when they exclude others from using it, and so on.
From the instant a joke is something that can accrue utility, like social prestige, and is
rival (you cannot tell the joke to the same people), the intuitive system triggers a clear
reaction, while the reflective ownership concept would have much difficulty explaining
why such contextual circumstances make all the difference. Obviously, it is possible to
refine our concept of ownership to accommodate such contextual cues. But that is not
required. As long as people share intuitions and have some confidence that they do,
the strategic norms are in place and transaction costs are diminished.
Given that coordination is made more efficient by explicit concepts and words,
what explains the discrepancy noted above, between these explicit understandings and
our intuitions? Note that the discrepancy is not specific to the domain of ownership.
Indeed, one can observe a similar phenomenon in most cognitive domains handled by
both intuitions and deliberate, explicit reflections, as described by dual process models
(see, e.g., Evans 2003). The fact that explicit representations do not always track intu-
itions is mostly due to computational differences in the two kinds of systems described
in such dual-process models.
Intuitions are delivered by nonconscious integration of many different types of
cues, each of which triggers inferences that are then weighted for relevance and trig-
ger specific intuitive content, as well as motivational states. That much is common
to many domain-specific inferential systems. In the case at hand, the relevant cues
include information about whether the resource is rival (which is itself the outcome of
previous computation, based on semantic knowledge but also on episodes or inferences
from episodes), information about the history of the connection between an agent
and a thing, information about other agents’ past interaction with the same thing,
and much more besides that. Because the cues are many, and because their inferential
194 Pascal Boyer

potential is often modulated by the values of other cues, they do not lend themselves
to the kind of inference typical of explicit, reflective systems. Explicit systems are con-
strained by the limits of working memory and imagery buffers, which is why they
cannot in general represent complex webs of contingencies between multiple aspects
of a situation. Explicit thoughts on ownership provide general representations that, to
some rough extent, track the usual results of intuitive processes concerning possession
and ownership, in the forms of general principles about ownership, for example, “what
is owned by X cannot be used by Y,” “a gift of R from X to Y entails that X does not
own R any longer,” and so on.

7.4.2 How Do We Extend the Scope of Ownership Concepts?

Human representations of ownership are not exhausted by intuitions and the kind
of explanatory, comment-like reflections described above. In fact, in many human
groups, people have put together formal codes of ownership, including legal systems,
that extend the scope of intuitive and reflective representations. Most important, these
formal systems apply ownership concepts to domains of social interaction for which
there are no evolved predispositions. For instance, humans have evolved intuitions
about territories and their use. If you occupy a territory, you can exclude others from
resources found in that space. But modern legal norms make the exploitation of min-
eral resources and the use of airspace possible and efficient, by creating specific limited
property rights, by creating such notions as royalties and refunds for negative externali-
ties, and so forth. Indeed, modern legal scholarship tends to deny that there is such
a thing as ownership in the abstract, which is replaced with the notion of bundled,
highly specific rights, such as the right to occupy land but not sell it, or the right to
sell property but not to damage it, and so forth. (Merrill and Smith 2001). By build-
ing these nonintuitive notions, legal systems can extend ownership to domains such
as ideas, tunes, and designs, for which our common reflective notion of ownership is

7.5 Implications of the Evolutionary Perspective

A naturalistic evolutionary perspective on mental content diverges in important ways

from standard assumptions about concepts. For a long time, theoretical debates about
the structure of concepts focused on categorization, notably on prototypicality effects,
the role of perceptual imagery, and artificial intelligence models of knowledge repre-
sentation (see, e.g., Gelman and Medin 1993; Medin, Goldstone, and Gentner 1993;
Medin and Wattenmaker 1987; Murphy 2002). Empirical evidence persuaded psychol-
ogists to move away from seeing concepts as lists of properties (the so-called classical
view), to recognizing that they include prototypes or exemplars, and finally to constru-
ing them as encapsulated theories about kinds of objects (Medin and Wattenmaker
How Natural Selection Shapes Conceptual Structure 195

1987). These debates have now lost much of their intensity, perhaps because of a
realization that the world of mental concepts is far more extensive than previously
envisaged, and that representational formats are far more diverse. Depending on the
domain, category structure is best described by one format or another, but none of
them applies across many different domains (Machery 2009). Moreover, a large part of
our conceptual menagerie includes ineffable, modality-specific information that does
not easily lend itself to a unified nonmodal code (Barsalou 1993, 1999). These findings
converge to suggest that there is simply no possibility of (or need for) a general theory
of concepts, if this means a series of general strictures on the representational format of
all or most concepts (Machery 2009).
But cognitive and evolutionary considerations may allow us to go further, explain-
ing why concepts come in diverse formats and why these are appropriate to specific
conceptual domains. As a starting point, an evolutionary perspective may introduce
important corrections to standard assumptions about concepts, to do with (a) the
connections between intuitive and reflective representations, (b) the connection
between concepts and valuation, and (c) the contribution of concepts to meaning and

7.5.1 Intuitions and Reflections

Concepts as described here include both intuitive components, whose inferential back-
ground is not available to conscious inspection, and reflective components, which
explicate, enlarge, modify, or otherwise elaborate on intuitions. Consider the example
of OWNERSHIP. This is involved in a large spectrum of situations, all the way from simple
tactical conflict avoidance (e.g., when three-year-olds tacitly abide by the first posses-
sion principle), to large-scale coordination (as when people elaborate property rights
for a community). In the first case, the selective pressure is for efficient strategic norms,
which require that roughly similar cues trigger roughly similar intuitions and motiva-
tions in conspecifics. In the case of large-scale interaction, the complexities of coordi-
nation make explicit verbalized rules an efficient addition to intuitions.
This interaction of intuitive and reflective components is familiar in many cogni-
tive domains and usually described as an example of dual-process systems (Evans 2003;
Hassin, Uleman, and Bargh 2005). From a selective viewpoint, one would expect pres-
sure toward fast, rough-and-ready systems in some domains and slower, integrative
systems in other domains. However, again based on these functional considerations,
one would certainly not expect the mind to comprise two general-purpose systems
that support gut-feeling and reasoning-driven decision making respectively, as is some-
times argued in the dual-process literature (Kahneman 2003). That is because decision
making itself, rather than being a unified, all-purpose mental faculty, is fragmented in
myriad distinct domain-specific capacities, with their distinct input formats, databases,
and computational rules (Barrett and Kurzban 2012).
196 Pascal Boyer

7.5.2 Concepts and Valuation

As standard research on concepts focused on categorization, it left aside aspects of
concepts formation and use that are just as crucial, and especially salient from an
evolutionary viewpoint. Concepts, like other cognitive structures, are there to direct
behavior efficiently in the context of specific selective pressures. So concepts should
be such that they result in appropriate motivations, in the different valuation of dis-
tinct courses of actions (Tooby, Cosmides, and Barrett 2005). This is clear in the case
of ownership. Indeed, two distinct situations may be construed as cases of ownership
versus mere possession, as a result of salient differences in the motivations triggered.
For instance, having created an artifact gives us a greater motivation to defend our
possession of the artifact than if we had just found it, and this difference in motivation
is what creates the intuition that the artifact “belongs” to us. It seems contrived and
artificial to distinguish categorization from motivation in this case.
The same may apply to many other concepts. As Tooby and colleagues point out,
valuation is ubiquitous in the representation of situations, such that a division between
categorization and motivation makes little biological sense (Tooby, Cosmides, and
Barrett 2005). Indeed, one may argue that the separation between conceptual functions
understood as pure categorization (describing which objects go together) and motiva-
tion understood as a separate process of decision making may be a hangover from the
classical faculty psychology description of volition. To take a simple example, the cog-
nitive system that makes a male chimp notice a female’s genital swelling, rather than
treat it as visual noise, is the very same system that triggers sexual arousal in the male
and prompts courtship behaviors. Or, to return to the dog’s ontology as mentioned in
introduction, the systems that notice differences between humans and other animate
beings are the same systems that motivate highly specific behaviors toward humans.
The point is that valuation is not some external factor that is added to categoriza-
tion. Valuation generates categories by motivating different behaviors toward different
objects (Tooby, Cosmides, and Barrett 2005).

7.5.3 Concepts and Reference: Deflationary Implications

To the extent that philosophers consider mental content to contain structured bundles
of information, they generally assume that these conceptual bundles should be such
that they can support reference (Block 1987; Rey 1983). There is no space here to dis-
cuss accounts of reference and their connections to various hypotheses about mental
content. In fact, a normative notion of reference perhaps has no place at all in a natu-
ralized account of cognition, as some philosophers have argued (Davidson 1984; Quine
1960). But it may be of help briefly to signal how a naturalistic evolutionary framework
diverges from common normative expectations about concepts.
A possible objection to an evolutionary perspective is that concepts construed in
relation to evolutionary pressures denote not genuine kinds in the world, but broader
How Natural Selection Shapes Conceptual Structure 197

equivalence classes of phenomena with similar effects on a lineage of organisms. For

instance, in this view, it is difficult to say that cows have a concept GRASS, when in fact
all they seem to have is the concept GREEN, GRASSLIKE LOOKING, SMELLING AND TASTING STUFF,
which fails to pick out all grass and only grass. This objection is not really compelling,
however, for two reasons. First, a failure to pick out genuine natural kinds in the world
is only a problem if one assumes that the point of concepts is to provide organisms
with scientifically accurate or metaphysically coherent accounts of the world. By con-
trast, once we consider concepts as information structures that help organisms survive
and reproduce, metaphysically imperfect concepts can be seen as perfectly fine tools.
In this perspective, cows and other ruminants have managed rather well so far with
GREEN, GRASSLIKE LOOKING, SMELLING AND TASTING STUFF mostly because, on average and over

eons of evolutionary history, there were not many nongrass objects in their environ-
ment that matched this “flawed” concept. Second, most of our mental concepts track
kinds of things that are not proper categories in any case. Even so-called natural kinds
are generally not actual classes (Millikan 2005, 106ff). Natural organisms come in dif-
ferent species and genera, that is, not in classes but in populations that merge with
other populations as one goes back in time and that diverge into distinct species as
time passes (Mayr 1996). So there is no distinct class of objects in the world that GRASS
or TIGER could normatively refer to. From a naturalist standpoint, then, concepts can do
their functional work without “metaphysical correctness.”
In functional terms, once we abandon normative metaphysical requirements, con-
cepts are best described as a set of specialized computational skills (Millikan 1998), with
specific triggers, a specific set of rules for operations, and a range of typical outputs,
for example, a modification of that information, the recruitment of specific informa-
tion from memory, a specific motivation or emotional reaction, and so on. The case
of OWNERSHIP illustrates this. The range of mental representations associated with pos-
session and use do not converge on a set of criteria, such that a particular agent could
be said to grasp the concepts OWN or BELONG while another one does not. There are
shared intuitions (e.g., about first possession), external cues that influence these intu-
itions (e.g., the way a resource was extracted from an environment), as well as reflec-
tive thoughts on the intuitions (e.g., about what in general determines ownership and
what it implies). The connections between actual situations and the activation of all
these representations is a matter of greater or lesser success in tracking relevant agent-
environment interactions.
More generally, what evolved organisms need are not information bundles that
connect to the way the world is, but information bundles that track the way the
world affects fitness, which is a property not of the world outside the organism, but
of the combination between that world and species-specific adaptations. Taking
this into account may allow us to make sense of properties of human concepts that
would otherwise remain puzzling and provides a way of mapping human concepts
198 Pascal Boyer

that is based not on a priori philosophical requirements, but on biological matters

of fact.


Bakeman, R., and J. Brownlee. 1982. Social rules governing object conflicts in toddlers and pre-
schoolers. In Peer Relations, ed. K. H. Rubin and H. S. Ross, 99–111. Hillsdale, NJ: Erlbaum.

Barrett, H. C., and R. Kurzban. 2012. What are the functions of System 2 modules? A reply to
Chiappe and Gardner. Theory & Psychology 22 (5): 683–688.

Barsalou, L. W. 1993. Flexibility, structure, and linguistic vagary in concepts: Manifestations of

a compositional system of perceptual symbols. In Theories of Memory, ed. A. F. Collins, S. E.
Gathercole, M. A. Conway, and P. E. Morris, 29–101. Hillsdale, NJ: Erlbaum.

Barsalou, L. W. 1999. Perceptual symbol systems. Behavioral and Brain Sciences 22 (4): 577–660.

Barsalou, L. W., W. K. Simmons, A. K. Barbey, and C. D. Wilson. 2003. Grounding conceptual

knowledge in modality-specific systems. Trends in Cognitive Sciences 7 (2): 84–91.

Blake, P. R., P. A. Ganea, and P. L. Harris. 2012. Possession is not always the law: With age, pre-
schoolers increasingly use verbal information to identify who owns what. Journal of Experimental
Child Psychology 113 (2): 259–272.

Blake, P. R., and P. L. Harris. 2009. Children’s understanding of ownership transfers. Cognitive
Development 24 (2): 133–145.

Block, N. 1987. Advertisement for a semantics for psychology. Midwest Studies in Philosophy 10
(1): 615–678.

Brown, D. E. 1991. Human Universals. New York: McGraw Hill.

Carey, S. 2009. The Origin of Concepts. New York: Oxford University Press.

Cosmides, L., and J. Tooby. 2000. Consider the source: The evolution of adaptations for
decoupling and metarepresentation. In Metarepresentations: A Multidisciplinary Perspective, ed.
D. Sperber, 53–115. New York: Oxford University Press.

Davidson, D. 1984. Inquiries into Truth and Interpretation. New York: Clarendon Press.

DeScioli, P., and B. J. Wilson. 2011. The territorial foundations of human property. Evolution and
Human Behavior 32 (5): 297–304.

Dubreuil, B. 2010. Paleolithic public goods games: Why human culture and cooperation did not
evolve in one step. Biology and Philosophy 25 (1): 53–73.

Eisenberg-Berg, N., R. J. Haake, and K. Bartlett. 1981. The effects of possession and ownership on
the sharing and proprietary behaviors of preschool children. Merrill-Palmer Quarterly 27 (1):
How Natural Selection Shapes Conceptual Structure 199

Evans, J. S. B. T. 2003. In two minds: Dual-process accounts of reasoning. Trends in Cognitive

Sciences 7 (10): 454–459.

Friedman, O. 2010. Necessary for possession: How people reason about the acquisition of owner-
ship. Personality and Social Psychology Bulletin 36 (9): 1161–1169.

Friedman, O., and K. R. Neary. 2008. Determining who owns what: Do children infer ownership
from first possession? Cognition 107 (3): 829–849.

Friedman, O., K. R. Neary, M. A. Defeyter, and S. L. Malcolm. 2011. Ownership and object
history. In Origins of Ownership of Property: New Directions for Child and Adolescent Development, vol.
132, ed. H. H. Ross and O. Friedman, 79–90. New York: Wiley.

Gat, A. 2006. War in Human Civilization. New York: Oxford University Press.

Gelman, S. A., and D. L. Medin. 1993. What’s so essential about essentialism? A different perspec-
tive on the interaction of perception, language, and conceptual knowledge. Cognitive Development
8 (2): 157–167.

Gendler, T. S. 2008. Alief in action (and reaction). Mind & Language 23 (5): 552–585.

Hann, C. M. 1998. Property Relations: Renewing the Anthropological Tradition. New York: Cambridge
University Press.

Hassin, R. R., J. S. Uleman, and J. A. Bargh, eds. 2005. The New Unconscious. Oxford: Oxford
University Press.

Heine, B. 1997. Possession: Cognitive Sources, Forces and Grammaticalization. Cambridge:

Cambridge University Press.

Hrdy, S. B. 2009. Mothers and Others: The Evolutionary Origins of Mutual Understanding. Cambridge,
MA: Belknap Press.

Jackendoff, R. 1983. Semantics and Cognition. Cambridge, MA: MIT Press.

Johnsson, J. I., F. Nöbbelin, and T. Bohlin. 1999. Territorial competition among wild brown trout
fry: Effects of ownership and body size. Journal of Fish Biology 54 (2): 469–472.

Kahneman, D. 2003. A perspective on judgment and choice: Mapping bounded rationality.

American Psychologist 58 (9): 697–720.

Kanngiesser, P., N. Gjersoe, and B. M. Hood. 2010. The effect of creative labor on property-
ownership transfer by preschool children and adults. Psychological Science 21 (9): 1236–1241.

Kim, S., and C. W. Kalish. 2009. Children’s ascriptions of property rights with changes of owner-
ship. Cognitive Development 24 (3): 322–336.

Machery, E. 2009. Doing Without Concepts. New York: Oxford University Press.

Margolis, E., and S. Laurence. 2007. The ontology of concepts—abstract objects or mental repre-
sentations? Noûs 41 (4): 561–593.
200 Pascal Boyer

Maynard Smith, J. 1982. Evolution and the Theory of Games. Cambridge: Cambridge University

Mayr, E. 1996. What is a species, and what is not. Philosophy of Science 63 (2): 262.

Medin, D. L., R. L. Goldstone, and D. Gentner. 1993. Respects for similarity. Psychological Review

Medin, D. L., and W. D. Wattenmaker. 1987. Category cohesiveness, theories and cognitive
archaeology. In Concepts and Conceptual Development, ed. U. Neisser, 25–62. Cambridge:
Cambridge University Press.

Merrill, T. W., and H. E. Smith. 2001. What happened to property in law and economics? Yale
Law Journal 111 (2): 357–398.

Millikan, R. G. 1998. A common structure for concepts of individuals, stuffs and real kinds: More
mama, more milk, more mouse. Behavioral and Brain Sciences 21:55–100.

Millikan, R. G. 2005. Language: A Biological Model. New York: Oxford University Press.

Murphy, G. L. 2002. The Big Book of Concepts. Cambridge, MA: MIT Press.

Noles, N. S., and F. Keil. 2011. Exploring ownership in a developmental context. In Origins
of Ownership of Property: New Directions for Child and Adolescent Development, vol. 132, ed.
H. H. Ross and O. Friedman, 91–103. New York: Wiley.

Palamar, M., D. T. Le, and O. Friedman. 2012. Acquiring ownership and the attribution of respon-
sibility. Cognition 124 (2): 201–208.

Quine, W. V. O. 1960. Word and Object. Cambridge, MA: MIT Press.

Rey, G. 1983. Concepts and stereotypes. Cognition 15 (1–3): 237–262.

Ridley, M. 2010. The Rational Optimist: How Prosperity Evolves. 1st U.S. ed. New York: Harper.

Ross, H. S., and C. L. Conant. 1992. The social structure of early conflict: Interaction, relation-
ships, and alliances. In Conflict in Child and Adolescent Development, ed. C. U. Shantz and
W. W. Hartup, 153–185. New York: Cambridge University Press.

Rudmin, F. W. 1994. Cross-cultural psycholinguistic field research: Verbs of ownership and

possession. Journal of Cross-Cultural Psychology 25 (1): 114–132.

Sperber, D. 1997. Intuitive and reflective beliefs. Mind & Language 12 (1): 17.

Sperber, D., and D. Wilson. 1998. The mapping between the mental and the public lexicon.
In Thought and Language, ed. P. Carruthers and J. Boucher, 184–200. Cambridge: Cambridge
University Press.

Tooby, J., L. Cosmides, and H. C. Barrett. 2005. Resolving the debate on innate ideas: Learnability
constraints and the evolved interpenetration of motivational and conceptual functions. In The
Innate Mind: Structure and Contents, ed. P. Carruthers, S. Laurence, and S. Stich, 305–337. New
York: Oxford University Press.
IV Concepts and Perception
8 Burge on Perception

Jerry A. Fodor

8.1 Introduction

Tyler Burge has written a long,1 disputatious, and sometimes difficult book about cog-
nition, with special emphasis on the relation between perception and conceptualiza-
tion (Burge 2010). Among its many other virtues, it is a serious work of scholarship.
People like me, who only rarely can manage to get their references to stay put, will be
awed by its fifty-odd pages of bibliography, which cites not just the usual standards in
the philosophy of mind and the philosophy of language, but also such exotica as, for
example: “development in young infants’ reasoning about the occlusion of objects,”
“exoskeletal sensors for walking,” and a lot of other books and papers drawn from the
professional literature of psychology and related sciences. It is a contention of Burge’s
that an embarrassing number of the theses that philosophers of mind have held to be
true a priori turn out, in light of the empirical findings, not to be true at all. (Burge’s
favorite examples are about what, in light of their lack of language, animals and prelin-
guistic infants can and can’t perceive.) Burge is surely right to consider that a scandal.
In the old days, philosophers thought that they could arrive at conceptual truths by
performing conceptual analyses; so it is perhaps understandable that, when they got
around to the philosophy of mind, they ignored empirical findings and did their stuff
from armchairs. But then Quine pointed out that successful conceptual analyses have
been remarkably thin on the ground, and that nobody seems to be able to say just what
a conceptual analysis is. The prospects for an analytic theory of perception now seem
about as promising as the prospects for an analytic theory of continental drift; and
much the same can be said of the prospects for analytic theories of believing, thinking,
remembering, and other cognitive states and processes.
I think that Burge’s methodological observations about how philosophical research
on perception and cognition should be conducted are patently correct; this paper will

1. Origins of Objectivity is approximately twenty times longer than the reference copy of Winnie-
the-Pooh that I keep on my desk to use in emergencies; and it contains many, many fewer jokes.
204 Jerry A. Fodor

take them for granted. My primary concern will be the question: “Given that theories
of perception/cognition2 must be responsible both to empirical findings and to reason-
able demands for conceptual coherence, conservatism, and the like, what theory of per-
ception—or, at a minimum, what kind of theory of perception—should we endorse?” Burge’s
book repeatedly asks philosophers, psychologists, linguists, ethologists, and anybody
else who is prepared to lend a hand, what is required of a creature, and of its environment,
such that the one should be able to perceive the other “objectively”?3 I think this is indeed
a sort of question that we should all be asking and that those are the sorts of people of
whom we should ask it. But I’ll argue that Burge gets the answer wrong—interestingly
wrong, but wrong all the same.
This chapter comes in sections. Section 8.2 is an overview of (what I take to be) the
dialectical landscape. This is necessary because Burge’s way of carving things up cuts
across a distinction that I take to be essential: perception versus perception as.

8.2 Overview

So what, according to Burge, does one discover if one addresses theories of percep-
tion with due attention to the available empirical results? Primarily that philosophers
have overintellectualized their accounts of how perception works (and, perhaps, their
accounts of how thought does). In particular, Burge thinks that philosophers have
done so by holding:

(i) perception typically requires conceptualizing the percept, so what one can perceive
depends on which concepts one has;


(ii) which concepts one has depends on which beliefs one holds.4

This, however, is where exegesis gets sticky. As far as I can tell, Burge thinks that he
thinks that (i) and (ii) are both false. I think, however, that his position with respect to

2. To conserve backslashes, I’ll use theory of perception as short for “theory of perception or
of other aspects of cognition” except when I think it matters which of the disjuncts is being
discussed. I assume (untendentiously, I hope) that perception is a branch of cognition, hence
that the science of perception is a branch of cognitive science. Much of this chapter is about
which branch of cognitive science it is and how that part differs from some of the others.
3. To see a thing “objectively” is, at very least, not to see it as having properties that it doesn’t
have. On Burge’s view, several features of perception are explained by their being conducive to
perceptual objectivity so construed. Burge holds that perceptual constancies are paradigms of
4. This sort of view of the semantic content of concepts is often called internalism, or individual-
ism. Burge uses both terms, as shall I in what follows.
Burge on Perception 205

(i) is, in fact, equivocal and that the equivocation matters a lot to what he says about
related issues. By contrast, his rejection of (ii) is entirely clear: it is the core of what
Burge calls “anti-Individualism.” I’ll start with (i).
I want to remind you of the traditional distinction between seeing a thing that is an
F and seeing a thing (whether or not it is an F) as an F. Both seeing and seeing as can,
of course, be instances of objective perception. Burge is perfectly aware of the see/see
as distinction, but he seems not to think it bears much weight. In fact, he sometimes
uses the two expressions more or less interchangeably. That, in my view, is a Very Bad
Mistake; and I think that Burge’s treatment of the relation between perception and
conceptualization depends on his making it.
An example: A revealing passage on page 244 is germane.5 Burge is arguing (rightly, I
believe) that in the paradigm cases, the object of sensation is the same as the object of
perception: What we sense, like what we see, is a cat lurking in the bushes (not a glimpse,
or a “sense datum,” or an array of shapes, colors, textures … and so forth as internalists
have often supposed). Burge remarks: “… one can immediately and fully apprehend
something as a body. … It is not evident that it takes extra time, beyond the moment
of viewing, to see a body, or to see something as a body.” This sounds to me like a
confusion of two claims: On the one hand, there’s the (as Wittgenstein might have put
it) “grammatical” observation that, if there really is a cat in the bushes, then seeing it
and sensing it may be the very same event. (They come apart, however, if there isn’t
a cat.) On the other hand, there’s the question whether seeing a cat as a cat “takes
longer” than (merely) seeing a cat. Only the second question bears on the issue whether
inferential (hence conceptual) processes mediate seeing. Just seeing a cat may require
some conceptualization, but it surely doesn’t require exercising (or even having) the
concept CAT. Consider the case where, though what you see is in fact a cat, what you
see it as is an old shoe. “See” is transparent to substitution of identicals. We typically
see things as things; but that doesn’t settle whether we typically see them as something
else first, as individualists have traditionally supposed.
Just what I think has gone wrong is a longish story, but a rough approximation fits
in a nutshell: on the one hand, there is no perceiving without perceiving as; and, on the
other hand, there is no perceiving as without conceptualization; and, on the third hand,
there is no conceptualization without concepts. That is what’s right about (i) and the
traditions in the philosophy of mind and in the psychology of perception that endorse
it. What’s wrong with that tradition is its forever wanting to infer from “perceiving
requires conceptualization” (because it requires perceiving as, and perceiving as is a
species of conceptualization) to “perception must be subject to epistemic constraints.”
The link between the premise (which is true) and the conclusion (which isn’t) is the
idea that there are typically epistemic constraints on concept possession, an idea that

5. All page references are to Burge (2010).

206 Jerry A. Fodor

I take to be false root and branch. Accordingly, the path of virtue is to concede that
perceiving requires conceptualizing, hence that (contrary to Burge) a viable theory of
perception presupposes a viable theory of concept possession. What should be denied
to the last ditch, however, is that a viable theory of concept possession requires epis-
temological foundations. Summary so far: Burge accuses conceptualists6 of having
overintellectualized the psychology of perception. But conceptualism is okay; seeing
something at all requires seeing it as something-or-other.
Burge is right to say, as he repeatedly does, that having the concept F does not
require knowing “what it is for something to be an F”; or “how to tell whether some-
thing is an F”; or “what it’s like to see an F” (still less “what it’s like to be an F”); or “what
the criteria for F-ness” are; or “what belongs to the F stereotype”; and so on through
many other epistemological theses about the conditions for concept possession that
philosophers and psychologists have championed from time to time. Thus, a concep-
tualist might consistently hold both that conceptual content supervenes (not on what
one believes but) on mind-world causal connections, and that perception requires con-
ceptualization. Which is to say that conceptualism and anti-individualism can both
be true. That is, after all, unsurprising since individualism is about the metaphysics of
conceptual content, not about the role of concepts in perceptual (or other) psychologi-
cal processes. Theories of mind can thus contrive to have the best of both worlds, but
only at a price: they are required to bear in mind that perceiving requires perceiving as
and that perceiving as requires conceptualization. But, because he holds that percep-
tion doesn’t imply conceptualization, Burge can’t do so. That’s the short form of the line
of argument I will pursue; details to follow.
Seeing something as an F requires seeing it as an F; by contrast, just seeing an F
does not since you can perfectly well see an F by seeing it as a G. For convenience of
exposition, philosophers like to say this sort of thing in the formal mode (thus invit-
ing the vulgar complaint that philosophy consists of quibbles about words): “perceives
an F that is G” is transparent at the G position, but “perceives an F as G” is opaque at
the G position. If you see a dog, and the dog you see is brown, then it follows that you
see a brown dog. That’s so whether or not you notice that the dog is brown, indeed,

6. By stipulation, a conceptualist is someone who holds that perception requires conceptualiza-

tion of the percept. On the conceptualist view, seeing a thing is rather like thinking about the
thing: both require the application of concepts to their objects. Conceptualists hold that “no
concepts, no thoughts” and “no concepts, no percepts” are both true: seeing something and
seeing something as F requires having the concept F. What’s wrong with the tradition Burge
opposes isn’t that it overintellectualizes perceiving; it’s that it epistemologizes concept possession.
It’s essential to distinguish between theories about the role of concepts in perceiving and theories
of concept possession. I’m going to argue that the conditions for concept possession are, in gen-
eral, not epistemic.
Burge on Perception 207

whether or not you notice that it is a dog. By contrast, to see a thing as a brown dog
is to see it as brown and as a dog; that is so whether or not it is, in fact, either. To see
a red geranium as a brown dog is to see a red geranium as brown and as a dog, even
though it is neither the one nor the other. This feature of “see as” (and other opaque
contexts) really is puzzling; there simply is no rug under which it can be swept. In the
case imagined, all there is in the world is you and a red geranium that you see, so how
does the brown dog come in?
The immensely plausible answer is that it comes in via the concepts deployed in the
seeing as. A cat may look at a queen and, in favorable conditions, may even see one.
But, though I hold cats in great esteem, I doubt that a cat can see a queen as a queen.
The reason it can’t is, surely, that doing so requires having the concept QUEEN, which
cats do not. At one point Burge remarks that a science of vision will need to “take
perceptual states as representational [sic] in a psychologically distinctive sense” (Burge
2010, 347). Quite so. But in what psychologically distinctive sense? Answer: taking a
perceptual state as representational requires a distinction between what is perceived and
what it is perceived as; and “see as” is an opaque context; and seeing something as F
requires deploying the concept F. Compare zoology; that a thing is a cat does not raise
the question “what it is a cat as?”
So then: On the one hand, no cat that lacks the concept QUEEN can see a queen
as such. But, on the other hand, a cat can’t see a queen without seeing her as some-
thing-or-other. All that being so, a cat can’t see anything at all if it has no concepts
at all. Cats without concepts are blind. (Didn’t Kant say something of that sort some
while ago?)
I don’t know what Burge would say about that; as I remarked, he seems not to make
much of the “see”/”see as” distinction. But it is no help in sorting this out that Burge
regularly uses versions of the locution “perception of x as of an F” both for perceiving
an x that is F and for perceiving an x as F, thereby eliding the distinction between them.
So, for example, when Burge is discussing the way that perceptual constancies militate
in favor of the objectivity of perception, he speaks of the perception of a round plate as
being “as of something round” even when the plate is tilted (383; I’ve slightly altered
Burge’s example, but not in a way that affects the present issues). But (in a passage
about the mental mechanisms that mediate the constancies) he says that “the trans-
formational story begins with two-dimensional retinal sensory registration and ends
with visual perceptions as of attributes in the physical environment” (358). This way
of talking raises a crucial question that Burge doesn’t answer: whether you can have
a perception “as of an attribute in the physical environment” if there is, in fact,
nothing in the physical environment that has that attribute; this is, in effect, the
question whether perception “as of an F” can have an opaque reading. That question
is crucial; we’ll presently come to why it is. First, a longish digression.
208 Jerry A. Fodor

8.3 A Longish, Largely Expository Digression

I think that, at several places where he shouldn’t, Burge relies heavily on an unexpli-
cated notion of “perception of an F,” thus begging questions that a theory of concep-
tual content ought to address head on. I’ll consider two examples.
First example: In the course of explaining what “anti-individualism” is, Burge
opposes the thesis that anti-individualism “is true of how referents are established, but
that some entirely different account is true of how referents are perceived or thought
about” (77). (The explicit target here is the theory that, in thoughts about water, water
is represented by a water stereotype.) Burge replies that “since anti-individualism con-
cerns ways of thinking as … its points cannot be captured by claiming that it concerns
only reference. … [Thinking of water] is a generic type of thinking that is absolutely not
to be identified with thinking of water as the colorless, odorless liquid that fills lakes
[etc.]” (79). Quite so. On the natural way of reading it, the thought that water is wet and
the thought that “[insert water stereotype] is wet” both make reference to water, but
they are nevertheless different thoughts; the difference between the two is thus exactly
analogous to the difference between perceives water and perceives as water. But (contrary
to what Burge appears to suppose) none of this even remotely suggests that the water
referred to in “perceives (/thinks about) water” is unconceptualized. Perhaps the differ-
ence between perceiving/thinking about water and perceiving/thinking about what satisfies
the water stereotype is that it is natural to read the first but not the second as transpar-
ent.7 In any case, as far as I can tell, Burge offers no construal of perceiving/thinking
about X, conceptualist or otherwise, except for his remark that it is “constitutively deter-
mined by patterns of interaction with the environment beyond the individual” (79),
which, since it is true of both perceiving (/thinking about) and of perceiving (/thinking
about) as, doesn’t distinguish the two. N.b.: the complaint here isn’t that Burge lacks a
reductive theory of perceiving (/thinking about); it’s that what he says about them doesn’t
bear on such crucial questions as, for example, whether perceiving (/thinking about)
Cicero as Tully counts as perceiving (/thinking about) Cicero; or whether perceiving
(/thinking about) water as H2O counts as perceiving (/thinking about) water.
To put it slightly differently, Burge’s way of construing anti-individualism makes it
compatible with holding that the veridicality condition on thinking about water impli-
cates external conditions only in the same boring way that the veridicality condition

7. It adds to the confusion that, like other opaque contexts, “thought about (/referred to) water”
is ambiguous: roughly, it is opaque to the substitution of identicals on the reading X thought
about (/referred to) water so described; but it is transparent to substitution of identicals on the read-
ing water is such that X thought about (/referred to) water. The point, in any case, isn’t really about
the logical form of English “think about/refer to” sentences. Rather, it’s that, on the preferred
readings, water is the intensional object of thoughts about water so described but not of thoughts
about water tout court.
Burge on Perception 209

on referring to water clearly does. It’s trivially true that what a thought refers to when
it refers to stuff in the world as water is stuff in the world (viz. water); likewise, what a
thought is about when it’s about water is also stuff in the world (viz. water). These are,
if you like, both “external” constraints on the content of water-thoughts; they both
constrain how the world must be in order for thoughts to represent stuff as water. But
that, surely, isn’t what anti-individualism is going on about when it says that content
is environmentally determined. If it were, individualists and anti-individualists could
very well lie down together since both agree that it’s water that water-thoughts repre-
sent; and that is itself an “external constraint” on the content of water-thoughts. Short
form: There is a robust sense in which for a thought to be about water isn’t (necessarily)
for it to be about water as such. Likewise, to perceive water isn’t (necessarily) to perceive
water as such. I take all that to be common ground. But it leaves wide open whether
perceiving (/thinking about) requires conceptualization (which is, of course, the main
issue that Burge is concerned about). The internalist/individualist says that perceiving
X requires perceiving X as something-or-other, just as thinking about X requires think-
ing about X as something-or-other. So, perceiving X and thinking about X are alike in
the crucial respect: both require conceptualization of X. That’s why philosophy and
psychology have traditionally embraced a conceptualist account of both perception
and thinking. Pace Burge, people hold that perception implies conceptualization not
because they overintellectualize perception; rather it’s because they have recognized
that perceiving (like thinking about) requires a “mode of presentation” of whatever it
is that is perceived. Both require concepts under which the object perceived (/thought
about) is subsumed by the perceiver (/thinker). Burge to the contrary notwithstanding,
perception really is a lot like thought. But whereas the conceptual mediation of thought is
self-evident, recognizing the conceptual mediation of perception was among the great
contributions of rationalist theories of mind. It took hard work both in the library and
in the laboratory to achieve. It is not to be dismissed lightly.
Second example: There is some rough going around pages 152–168 that I think traces
to Burge’s not having a robust notion of seeing as. Consider a case where a distal orange-
colored orange is displayed in a green light. A subject duly reports seeing a green(ish)
orange. I would think the natural way to describe this situation is that an orange orange
is being seen as green, hence as a situation in which an orange orange is misperceived.
This is just the sort of work that seeing as is good at doing. But (to repeat) Burge doesn’t
have a notion of seeing as; so what he says instead is that the representational content
of the subject’s perceptual state should be specified relative to the proximal stimulus,
not relative to the distal stimulus. Since, as we may assume, the proximal stimulus
when you see an orange orange in a green light is the same as the proximal stimulus
when you see a green orange in a white light, Burge says that the content of the per-
ceptual state that the subject is in when seeing an orange orange in green light is “as
of” seeing a green orange. So, to see an orange orange in green light as green isn’t, after
210 Jerry A. Fodor

all, to misperceive the orange. It counts as veridical perception according to Burge’s way
of counting.
But this is surely counterintuitive; and it’s a strange thing for Burge, of all people,
to say. Whatever else he is, Burge is a committed externalist; and an externalist
should specify perceptual content relative to its distal cause. To see something that
looks just like water as water is to misperceive it unless the distal stimulus is water.
And, anyhow, if seeing an orange orange in green light as green is veridical, what
on earth is it veridical of? Not the distal stimulus since, by assumption, the distal
stimulus is orange. Not the proximal retinal array because (unless you are very oddly
wired), you can’t see your retina without a mirror. Not your brain, because brains are
gray (more or less). Not an intensional object that is green, or a green sense datum, or
anything of that sort; Burge has no truck with any of those because he holds (rightly,
I think) that whatever seeing as is, it’s not a relation between a perceiver and a queer
object. Indeed, it’s not a relation between a perceiver and a mental object. So what,
then, is it?
Burge says that his way of understanding this sort of case is not a matter of philo-
sophical interpretation but “of scientific fact” (388). I suspect, however, that it’s really
an artifact of his failure to appreciate the extent to which psychology needs some
notion of seeing as even to describe the phenomena that it is trying to explain. It’s an
orange (not a sense datum or a proximal array that one sees as green when one sees
an orange in a green light). If that makes problems for semantics, or epistemology, or
ontology (or all three at once), such problems will just have to be faced and coped with.
Summary to here: “See,” “perceive,” and the like are typically transparent in the sense
that if you see/perceive an x, there is an x that you see. By contrast, “see as” and the
like are typically opaque; you can see something as an F even though no F is in view.
(Indeed, you can see something as an F if even if there isn’t anything that you see/
perceive as an F; as when, in a hallucination, you see the surround as populated by
elves.) No news so far.
Suppose, then, that you can have a perception of a plate “as of something round”
even though the plate is square. Then, to put it crudely, a conceptualist will want to
know where does the roundness come from? Well, where does it come from? Not from the
plate since, by assumption, the plate is square; but, if Burge is right, not from the plate
seen (as one says) “under the concept” ROUND, because, according to Burge, perception
doesn’t involve conceptualization. This is essential to his case; indeed, Burge’s argument
that the philosophy of mind is hyperintellectual consists mostly of his claim that phi-
losophers hold that perception is a species of conceptualization, which he thinks that
it isn’t. By contrast, the point of the conceptualist’s traditional arguments from per-
ceptual illusions, perceptual errors, and the like is that when a square plate is seen as
round, the “round” has to come from somewhere; and (short of “disjunctivism,” which
neither Burge nor I take very seriously) the only place that it could come from is a con-
cept that the perceiver applies to the plate.
Burge on Perception 211

Still, it may be that perception doesn’t need much by way of conceptual sophistica-
tion; externalists may be almost right in saying that what seeing needs is mostly that
the thing that’s seen is causally responsible (“in the right way”) for the seeing. But,
that won’t do except as a first approximation much in want of refinement. There is
no seeing a thing without seeing it as something-or-other, and when something looks
to you to be some-way-or-other, you must have some-or-other concept that it looks
to you to fall under; that seems plausible, surely, whether or not your preferred ontol-
ogy is committed to such things as looks. So, the chain of inference goes like this: no
seeing without seeing as; no seeing as without conceptualization; no conceptualization
without concepts. Thus do intentionality and conceptualization enter into the psy-
chology of perception; and they enter very early, certainly not later than perceptual errors
and illusions do. To that extent at least, the “argument from illusion” is vindicated, and
with it the conceptualist tradition. I don’t say that’s all self-evident; things really are a
mess in this part of the wood. But if, when you see a cat that is moving in the bushes
over there, you don’t see it as something moving in the bushes over there, why on
earth do you point your head toward the bushes when you want to see it better?
“But isn’t that just another instance of the very hyperintellectualization that Burge
complains about? Plants don’t turn towards the sun because they see it as the sun or
as anything else. Plants don’t have percepts; they have tropisms.” Burge makes this sort
of point again and again by way of deflating intellectualist accounts of perception. But
it’s really not relevant. Unlike plants, we do see things (and so surely do infants and cats),
and our having seen a thing often enters into the explanation of how we behave when
we see it. Our having seen a cat moving in the bushes may well be part of the story about
why we turned toward the bushes; but only if the story includes our having seen the cat as
something moving in the bushes. What enters into the explanation of our behavior isn’t the
objective stimulus (the stimulus that we actually see); it’s the effective stimulus (the objec-
tive stimulus as we take it to be). When the actual and the effective stimulus diverge, the
latter prevails—it controls behavior—all else equal. The constancies are there to assure
that the two don’t diverge very often in ecologically normal situations.
Take seeing as away and psychology loses the effective stimulus; take the effective
stimulus away and psychology loses the junction between perception and action. All
right so far, except that, by definition, when the effective stimulus isn’t actual, it isn’t
actually there; and, it’s presumably common ground that what isn’t there doesn’t cause
things to happen. The obvious way out of this is to say that, when the two diverge,
the properties of the effective stimulus are contributed not by the objective stimulus,
but by the perceiver’s conceptualization of the objective stimulus, that is, the objective
stimulus conceptualized (as “mentally represented”) by the perceiver. So, Burge to the
contrary notwithstanding, perception enters into the explanation of behavior only via
the assumption that percepts are conceptualized.
I don’t know how Burge proposes to get around that. Gibsonians try to do so by
speaking not of a creature’s response to the effective stimulus but of its response to
212 Jerry A. Fodor

what the (objective) stimulus “affords.” The improvement is, however, merely appar-
ent since what a certain stimulus affords to a perceiver depends not only on how the
stimulus actually is but also on what conceptual resources the creature brings to bear in
the perceptual process. So we’re back where we started; perception can explain behav-
ior only on the assumption that the percept is conceptualized.
So much for a priori argument; but here empirical considerations are also germane
since they strongly suggest, at the very least, that monkeys and infants do sometimes
see things as such-and-suches even when the things they see aren’t such-and-suches. For
example, there is evidence that babies and monkeys are sensitive to the ambiguity of
bi-stable figures like the Necker cube (see, e.g., von der Heydt, Qui, and Endo 2001).
This is relevant not only because it involves seeing two-dimensional figures as three-
dimensional, but also because there is reason to suppose that seeing a Necker cube
“switch” requires seeing some of the line intersections as line intersections on one
interpretation and seeing them as junctures of intersecting sides on the other. (This
is, I believe, the account of such illusions that perceptual psychologists usually offer;
Pylyshyn 2003.) Likewise, there is evidence that monkeys see figures in “Escher” draw-
ings as impossible objects(!) (Shuwairi, Albert, and Johnson 2007). So it would seem,
at a minimum, that Burge’s anticonceptualist claims about the concepts available to
the perceptual processes of baby’s and infra-human creatures argue much ahead of the
relevant science.8
Why, then, is Burge so set against conceptualist accounts of perception? There are
places where he seems to argue that for a mental representation really to be a concept,
it must function as a constituent of “propositional” thoughts. I suspect that Burge
thinks that the constituents of one’s thoughts just are one’s concepts. So if (as Burge
more or less takes for granted), infants/cats haven’t got propositional thoughts, it then
follows that they don’t have concepts either.
But, as far as I can tell, Burge doesn’t actually argue for any such tight connection
between having concepts and having propositional thoughts of which the concepts
are constituents. He therefore begs the question against a quite standard version of
modularity theory, according to which the concepts available in modular systems may
be encapsulated. (By definition, encapsulated concepts can’t interact with thinking or
any other “higher” cognitive processes). Perhaps the right thing to say is that concepts
are, of necessity, the kinds of things that can be the constituents of thoughts; but

8. Philosophers sometimes suggest that perception is direct (that is, not conceptually mediated)
when it is veridical. But they can get away with that only if they are prepared to assume what is
patently preposterous: that the psychological mechanisms that mediate veridical perception are
ipso facto different from the psychological mechanisms that mediate illusory perception. If the
reply is that the philosophy of mind doesn’t care about mediating psychological mechanisms,
the reply is that it bloody well ought to.
Burge on Perception 213

whether, in a given sort of creature, they do actually perform that function depends
upon the extent to which its cognitive architecture is modular (and, for all I know, on
lots of other things too). In any case, Burge needs an argument why only constituents
of actual propositional thoughts can be concepts; why, in particular, shouldn’t mental
representations that would be constituents of thoughts but for the constraints that a
modular cognitive architecture imposes be good enough to qualify as concepts? And
if that would be good enough, why shouldn’t such concepts have a role to play in the
perceptual integration of sensory registrations? This seems to be a perfect paradigm
of a question that can’t be settled from the armchair. To the best of my knowledge, as the
empirical findings now stand, it’s wide open that perceptual processes are often both
encapsulated and conceptualized. If I had to make a bet, that’s certainly what I’d bet on.9
You may think that that is the end. But it’s not.

8.4 Darwin Bashing

Rather surprisingly, the question we’ve been discussing—how should psychology

understand the distinction between seeing and seeing as?—bears interesting analogies
to a problem that seems, first blush, to live in a quite different part of the woods:
How should evolutionary biology understand the distinction between selection (as in
“homo sapiens was selected, but Neanderthal became extinct”) and selection for (as
in “having an opposed thumb was selected for in monkeys because having opposed
thumbs helped them to grasp things”). I want to say a bit about this, because it bears
on Burge’s treatment of the content of mental representations.
To begin with a couple of terminological points: Standard biological usage generally
has it that the objects of selection are kinds of creatures, typically species. So, the monkey
was selected, but the dodo was not. (Perhaps the dodo was selected against; or perhaps
it just died out.) By contrast, the objects of selection for are traits that are innate in a
species (their phenotypic traits). The core of the theory of natural selection (TNS)—the
thesis that is common ground among Darwinists as such—is that phenotypic traits are
ones that contribute to the adaptedness/fitness of the species that have them.10
So the motto is: In the paradigm cases, the fitness of creatures’ phenotypic traits explains
why they have the phenotypes that they do; which is to say that it’s the fitness of their

9. Oddly enough, Burge makes a quite similar point in an argument against Spelke (cf. 439).
Spelke contrasts two hypotheses about the mental representation of bodies: according to one, it is
intramodal; according to the other, it is transmodal. Burge replies, “why not both?” This question
is entirely pertinent; but it would seem to apply whenever issues about relations between percep-
tion and conceptualization arise.
10. More precisely, they’re traits that contributed to the adaptedness/fitness of members of the
species in the ecology in which the species evolved.
214 Jerry A. Fodor

phenotypic traits that explains why there are the kinds of creatures that there are. This
claim is, by general consensus, the heart of TNS.11
But here’s a point of logical syntax that really does matter and that Darwinists have
very largely been confused over: “select” is transparent (like “see”); but “select for” is
opaque, like “see as.” If an F is selected, and if being F is coextensive with being G, then
the selection of an F is ipso facto the selection of a G. By contrast, even if being F and
being G are coextensive, it does not follow that selection for being F is coextensive with
selection for being G (unless, of course being F and being G happen to be the same traits).
So, then, suppose two phenotypic traits are so linked that a creature that has either
has both. It would not follow that if either of these traits was selected for, so too was the
other. The reason that it wouldn’t follow is that the Darwinists hold it as an empirical
thesis that selection is for traits that contribute to adaptedness; and it’s easily imagined
that, of a pair of linked traits, one is adaptive and the other is neutral, or even that
having the other tends to reduce a creature’s adaptedness. (By the way, it isn’t just imag-
inable; it’s a sort of thing that happens all the time. For examples and discussion, see
Fodor and Piattelli-Palmarini 2010.) So, because “select for” is opaque to the substitu-
tion of coextensives, it’s perfectly possible that a creature has some phenotypic traits
that were selected for contributing to fitness and some that were not. But the central
thesis of TNS is that selection of creatures is selection for the adaptivity of their pheno-
types. So a situation of the kind just imagined would ipso facto be a counterexample to
TNS. That makes TNS a shaky foundation on which to build an account of conceptual
content, as we’re about to see.12

11. It’s sometimes said to be true by definition that the traits that are selected for are fitness
producing; but that can’t be right, and people really should stop saying it. For one thing, the
biologists’ consensus is that the measure of the fitness of a trait is not whether it is selected for but
rather the probability of its being inherited in successor phenotypes. This would make no sense
unless it is conceptually possible that some unfit traits are selected for and some fit traits are not.
And, more important for present purposes, TNS is the claim that having been selected for its
contribution to adaptedness explains why a trait belongs to a phenotype. But that couldn’t be so if
the connection between being phenotypic and being selected for were conceptual. John’s being
unmarried doesn’t explain his being a bachelor; John’s being unmarried is his being a bachelor.
12. Darwin was himself aware of the phenomenon of linkage; it shows up in breeding, where the
phenotype that you breed for is often not the one that you get. But he failed to notice that “select
for” like “breed for,” is intensional, or that its being so undermines the general claim that pheno-
types are selected for their adaptivity. For example, it might be that the posssible phenotypes are
constrained “from below” (e.g., by facts of biochemistry), and, very likely, “from above” as well
(e.g., by “laws of form”). See Fodor and Piattelli-Palmarini (2010) for extensive discussion of some
plausible candidates for such constraints. In our view, linkage is not an odd exception to TNS but
a salient example of how properties of a creature’s internal structure (i.e., non-ecological variables)
can shape the fixation of phenotypes.
That this line of thought was, and remains, widely misunderstood by the book’s critics argues
against the critics, not against the book.
Burge on Perception 215

Now back to Burge.

Burge very much needs a notion of biological function as part of his story about the
content of mental states because biological function is one of the things that Burge
appeals to when he wants to distinguish adventitious causes of such states from ones
that are, in his term, “constitutive of” their content. It turns out, for example, that
maybe you can satisfy the constitutive conditions for thinking about water at second
hand (indeed at umpteenth hand) since it might be sufficient for your thoughts to be
about water that some creature you evolved from had thoughts whose function was to
represent water in thought.
So, Burge thinks that an appeal to teleology is essential to making sense of
(mental) representation. But I’m not quite certain what account of teleology he
endorses. Through most of the book, the view seems to be pretty straightforwardly
Darwinian: to a first approximation, phenotypic traits, mental or otherwise, are
selected for their contribution to the fitness of individuals (or, perhaps of species);
and the function of a phenotypic trait is to do whatever it was selected for doing.
Accordingly, there is, as you would expect, much talk about the “basic” biologi-
cal needs of creatures, and the roles of their phenotypes in securing the satisfac-
tion of such needs. However, there are also places where Burge suggests that traits
can function to augment a creature’s “flourishing,” and I’m unclear how this more
or less Aristotelian notion of function is supposed to fit with the Darwinian kind,
where phenotypic traits are selected for their adaptivity. What does seem clear is
that “flourishing” can’t explain what selection for fitness purports to explain: viz.
why there are the phenotypes that there are. Not unless flourishing somehow reliably
predicts selection in the sense that TNS has in mind. For what it’s worth, though
it’s clear enough that many creatures manage to survive, it’s far from obvious that
there’s a lot of flourishing around. Lots of us are pretty depressed; and, for all I
know, lots of dinosaurs were too. In what follows, I’ll assume that Burge’s notion of
teleology is Darwinian, more or less.
Since their contents are constitutive of the identity of mental states, and since coex-
tensive mental states can differ in their contents, Burge needs a story about how the
process that mediates selection for fitness distinguishes among coextensive mental
states that differ in content. Here’s a more or less standard example of the problem:
Frogs are philosophically famous for snapping at flies, which they ingest whenever
they can. It’s thus natural to say that flies are the “intensional objects” of fly snaps; flies
are what frogs snap at. But what excludes the alternative thesis that it’s little ambient
black dots that frogs snap at? That is, what excludes the thesis that little ambient black
dots are what frogs see the flies they snap at as. On this account, ambient black dots are
what frogs like to eat, and catching them is what frogs have in mind when they snap
at flies. In an ecology where the ambient black dots are generally flies and vice versa,
a phenotypic disposition to snap at either would be equally good to satisfy the frog’s
appetite. So, then, which disposition do frogs have?
216 Jerry A. Fodor

These kinds of puzzles about mental content are called “disjunction problems” by
philosophers whom they trouble. Burge thinks that they are frivolous because they
ignore the function of the traits that constitute a creature’s phenotype. In the present
case, frogs snap at flies because doing so provides them with dinner, which is among
the frog’s basic needs. So, a Darwinian sort of account of the evolution of fly-snapping
behavior solves its disjunction problem, but only if relevant facts about the frog’s
ecology are borne in mind. So: the notion of selection for provides a notion of biological
function which, in turn, fills the gap between mental states that are merely coextensive,
on the one hand, and identical mental states, on the other. Putative instances of the
disjunction problem are artifacts of the characteristic philosophic mistake of ignoring
the empirical facts. Given TNS, there are no disjunction problems, so long as all of the
candidates are required to meet the test of what Gibsonians call “ecological validity.”
So I understand Burge to claim.
But that won’t do. Here as often elsewhere, Burge begs the issue by ignoring the
intensionality of the key notions; in particular, the intensionality of “select for” and
“needs.” As previously remarked, since it’s true (by assumption) that the ambient black
dots generally are flies (and vice versa) in the frog’s ecology, a snap at either will serve
a frog’s needs (and augment its flourishing) just as well as a snap at the other. In the
ecology they inhabit, frogs can get flies to eat by snapping at ambient black dots. So the
question “what does the frog see the thing it snaps at as” remains wide open, even if you
attend to the frog’s basic needs, as Burge suggests that you should. Likewise, of course, the
question “what function was the frog’s snap selected for performing?” See how high
the tides of intensionality run.13
It is, in short, trivial to provide truth conditions for “creatures of type X need …”
and for “phenotypic trait T is selected for …” when TNS is assumed and “need” and
“select for” are construed transparently. The problem is to provide truth conditions in
cases where “need” and “select for” are construed opaquely; it’s the opaque construal
of such terms that connects issues about what frogs need/want/snap at with questions
about how they mentally represent what they need/want/snap at. As far as I can tell,
however, the sort of account of mental content that Burge offers provides no slightest

13. I think that appealing to ecological validity is itself question begging in this context. Saying
that snapping at flies is what’s “ecologically valid” and snapping at ambient black dots isn’t, is
just a way of saying that (all else equal) the frog snap wouldn’t have evolved in nearby worlds where
there aren’t flies, but would have evolved in (some) nearby worlds where there aren’t ambient black dots
(perhaps, in those worlds, flies are red). So the appeal to ecological validity just replaces the
disjunction worry with the worry, “How could the process of selection distinguish between
mental states that differ in extension only in possible but nonactual worlds?” After all, only actual
causes can actually have effects. See, once again, Fodor and Piattelli-Palmarini (2010), where such
matters are discussed at length. (Fat lot of good that did us.)
Burge on Perception 217

clue of what the difference in mental content is between frogs that want to eat some
flies for dinner and frogs that want to eat some ambient black dots for dinner.14
Short form: It’s primarily in discussing “disjunction problems” that Burge uses
“perceive as” explicitly and extensively. That’s not surprising since a very natural way
of phrasing a disjunction problem is, “What do frogs see their prey as?” Burge likes
the idea that such questions are answered (or, perhaps, dismissed) by appeals to psy-
chological theories together with ecological facts of the sort pursued by investigations
in macrobiology, physiology, zoology, and related empirical sciences. But that doesn’t
help; it just moves the problems about intensionality from one guy’s desk to another’s.
“See as” is intensional; but so too are “select for,” “function as,” and “explain.” The
question “what do frogs snap at?” arises, sooner or later, in every science whose domain
contains intensional systems. (Gibsonians take note: “affords” is intensional too). So,
what do ambient flies “afford” to frogs? Food? Ambient flies? Dinner? Instantiated

8.5 Conclusion

There are lots of morals to be drawn from the discussion we’ve been having. One is
that it’s a bad idea to beg questions about intensionality; another is that it’s a good
idea, even for philosophers, not to ignore matters of empirical fact; another is that
Darwin didn’t, after all, solve the problems about the empirical underdetermination of
content. I want to emphasize the first of these in rounding off the chapter.
Philosophers have, by and large, taken it for granted that intensionality is, par excel-
lence, a topic in the theory of mind (or perhaps in modal logic, which some have hoped
might solve the problem in the theory of mind). There have been various suggestions
about what psychology should do about intensionality: one explicitly recommends
that it should give up on the mind and make do with just the material brain. Another is
that it should give up on matter and make do with just the immaterial mind. Another
is that it should give up on both matter and mind since intensional explanation
and neurological explanation are both just “stances” and are thus merely “optional.”

14. This is, of course, not just an argument against Burge’s treatment of mental content, but also
against any Darwinian account of biological teleology. Biologists have told me, from time to
time, that TNS must be true because biological explanation must have access to a notion of bio-
logical function, and only TNS can provide one. I don’t know whether it’s true that biological
explanation presupposes teleology (though Burge clearly thinks that psychological explanation
does); but whether it’s true that biology requires an account of teleology is beside the present
issues since TNS doesn’t provide one. Or rather, it does so only by playing on the intensional
ambiguity of “need,” “snap at,” “select for” (and, for that matter, “explain”), which nobody
ought to be allowed to do. Not even Darwin.
218 Jerry A. Fodor

And so forth. But whichever story is preferred, the view is that only psychologists and
philosophers need to worry over the problems that intensional explanation would
appear to raise; and, on second thought, maybe psychologists needn’t bother.
In the event, however, that view has proved to be wildly optimistic. Issues about
intensionality turn up all over the place, not just in psychological explanations insofar
as they are teleological, but also in biological explanations insofar as they rely on such
notions as selection for. And in historical explanations; and in explanations in econom-
ics; and in political science; and God only knows where else we would find them if we
chose to look. The moral is: It’s long overdue that we try to understand, in some detail,
just how intensional explanation works and, in particular, when it is appropriate in
science and when it isn’t. My objection to Burge’s book is largely that, by eliding the
distinction between see and the like and see as and the like, it distracts attention from
that project.
I don’t know where such a serious empirical account of intensionality might lead
(though I suspect that, at a minimum, major revisions are going to be required in theo-
ries about the fixation of phenotypes, TNS included). Suffice it that if, as Burge rightly
urges, it would be a good idea for philosophers to pay attention to relevant findings in
the sciences, it would likewise be a good idea for scientists to pay attention to relevant
findings in philosophy. Just for a change.

Appendix: The Doors of Perception, the Black Swan, and the Pale Cast of Thought

If I read it right, Burge’s book is committed to some or other version of this argument:
Burge’s argument:
1. Concepts are constituents of thoughts.
2. Infants and animals, though they don’t think (don’t have “propositional thoughts”), patently
do perceive.
3. Perception does not require the application (or even the possession) of concepts; in particular,
it shouldn’t be viewed, Kant-wise, as the integration of a sensory manifold under a concept.

By contrast, the present chapter is committed to some or other version of this

4. Seeing requires seeing as.
5. Seeing as requires conceptualization.
6. There is no seeing without conceptualizing.

This does appear to be a dilemma: at least one of these arguments must be faulty.
Which one?
I think it’s Burge’s. Because:

• 1 equivocates in a way that is important. Does it say that any concept a creature has
must be a constituent of some of its thoughts? Or does it say that any concept it has
Burge on Perception 219

can be a constituent of some of its thoughts (in effect, that concepts are the kinds of
things that can be constituents of thoughts)? Notice that although the former is incom-
patible with the conjunction of 1 and 2, the latter is not.
I’m inclined to think that it is some sort of a priori truth that concepts are ipso facto
potentially constituents of thoughts; but it’s not true a priori (in fact, it’s very likely
just not true) that a creature’s concepts are ipso facto constituents of some thoughts
that it actually has. There are many reasons to think that it’s not. For example, it seems
perfectly possible that a mind should have encapsulated concepts, including concepts
that are available for integrating sensory manifolds but not for thinking. There are, in
fact, empirical reasons to suppose that sort of thing happens quite a lot.
Assume, as psychologists often do (and as practically all the linguists I know actu-
ally believe) that there is a “language module” and that the utterances a speaker/hearer
understands are mentally represented within this module by parsing trees whose nodes
are labeled S, NP, VP, PP, and the like. Assuming that would not require holding that
a five-year-old who understands “that’s a large cat” ever does, or even can, think the
thought “large cat” is a noun phrase; it wouldn’t imply that five-year-olds ever think
anything at all about noun phrases. All that would follow is that children could think
thoughts about NPs if they could think all the thoughts that their repertoire of con-
cepts permits, which is exactly what modular accounts of mental processes deny. Burge,
who very commendably disapproves of flying in the face of science if one can avoid so
doing, would be the last philosopher to accept a concept of concept that makes genera-
tive grammar false a priori.
• The empirical status of 2 is very far from clear. Here’s a finding that animal psychol-
ogists routinely demonstrate to their introductory classes: In the experiment, a rat
(pigeon, whatever) is “reinforced for” pressing a bar. However, things are so arranged
that a bar-press eventuates in food only when it is made in the presence of a certain “dis-
criminative stimulus,” for example, only when a certain light is on. The empirical facts
are not in dispute.
The following explanation is prima facie plausible: The rat learns that the rein-
forcement of a bar press is contingent on the presence of the discriminative stimulus.
Having learned that, it reasons as follows: A bar press is reinforced if and only if the
light is on; on this trial, the light is (/isn’t) on; so on this trial I should (/shouldn’t) press
the bar. It proceeds to do so (/not do so) accordingly. Short form: the patterning of the
rat’s behavior is caused by its reasoning in accordance with modus ponens.
No doubt, that sort of explanation could be wrong. Perhaps the rat is a behaviorist
and thus thinks that rats don’t think; or perhaps it is an associationist, and thus thinks
that learned behaviors are associations of stimuli with responses. But, so far at least,
no remotely plausible behavioristic or associationistic explanation of the learning of
discriminative responses is on offer; and the question arises why, if rats shouldn’t be
220 Jerry A. Fodor

supposed to reason, it should be supposed that people do? But if rats can reason, surely
infants can too. Consider, for example, the recent experimentations on “number con-
stancy” in babies (reviewed in Carey 2009).

• Burge’s conclusion directly contradicts the view—prominent in the psychology of

visual perception—that many illusions (the Necker cube, the Müller-Lyer, the moon
illusion, among lots of others) turn on what the constituents of the stimulus are seen
as; according to one familiar account, to get the Müller-Lyer illusion, you must see
the figures either as corners projecting toward you or as corners projecting away from
you. It thus appears that an opaque notion of “seeing as” is sine qua non for a viable
theory of perception. (Analogous examples can easily be found in the literatures on
auditory perception and elsewhere.)
One might, I suppose, reply by claiming that 4 is in doubt. I don’t believe that it
is because I can’t think of a remotely plausible counterexample. Stendhal tells us that
del Dongo saw Napoleon at Waterloo, but only as a figure on a horse galloping by. He
might instead have seen him only as a spot on the horizon and he would have seen
Napoleon in either case; “see” is very transparent. But, if del Dongo did see Napoleon,
he must have seen him as something-or-other. Mustn’t he? If del Dongo didn’t see
Napoleon as something-or-other, how could he have seen Napoleon at all?

But I agree that that those are all pretty negative sorts of arguments. Here’s one that’s
less so: Perception can affect beliefs; presumably doing so is part of its job description.
Suppose you believe (wrongly) that all swans are white; and suppose you see a black
swan. Then (assuming rationality) shouldn’t the strength of your belief that all swans
are white be shaken? Well, no; it depends on whether you see the black swan as a black
swan. Perhaps you see it as a black channel marker (worse things happen at sea). Then,
all else equal, your seeing the black swan won’t (and oughtn’t) affect the strength of
your belief that all swans are white. “See as” is very much less transparent than “see.”
I conclude that the conceptualization of perceptual representations is prior to their
effects on the fixation of beliefs; n.b.: prior, even, to their effects on the fixation of
perceptual beliefs.
Versions of this line of argument have been in the philosophical literature at least
since Sellars, who thought that it refutes claims for a perceptual “given” of the kind
that sense-data theories endorse. But, in fact, it doesn’t; Sellars apparently failed to
consider the possibility that perceptual integration proceeds through several stages of
mental representation, and that a sense-data analysis might be true of some of them
even though it isn’t true of all of them. This is essentially the view endorsed by compu-
tational theories of perception according to which the earliest mental representations
of a percept are the outputs of “transducers.”
Burge on Perception 221

Burge has, however, a reply to “black swan” arguments. “I accept Sellars’s (Kant’s)
view that reasons that support knowledge must be propositional. … I do not accept
Sellars’s assumption that reasons are the only source of epistemic warrant” (435). But
that reply won’t do for the present purposes. The issue isn’t whether an account of
perceptual information as unconceptualized would be adequate to the purposes of a
normative epistemology; it’s whether such an account would be adequate to the pur-
poses of an empirical psychology. Normative issues entirely to one aside, if Burge is
right and belief content is conceptualized but perceptual content is not, how is it pos-
sible that creatures learn from their perceptual experience? Seeing a black swan as a black
swan would, really and truly, shake my belief that all swans are white. My belief that all
swans are white surely counts as propositional if anything does; modulo well-known
worries about generic NPs, it’s logical form is (something like) (x) x is a swan → x is
white. But then, how could I take my seeing a black swan as black to be counterinstance
to my belief that all swans are white unless I mentally represent the content of my
perception as (something like) ∃x (x is a swan and not white)? How else could I learn from
my perceptual experience that all swans are white isn’t true?
If the mental representations that express the objects of belief have conceptual
structure, then so too do the mental representations that express the objects of percep-
tion. The (dis)confirmation of beliefs by perceptions won’t settle for less.


Burge, T. 2010. Origins of Objectivity. New York: Oxford University Press.

Carey, S. 2009. The Origins of Concepts. New York: Oxford University Press.

Fodor, J., and M. Piattelli-Palmarini. 2010. What Darwin Got Wrong. New York: Farrar, Straus and

Pylyshyn, Z. 2003. Seeing and Visualizing: It’s Not What You Think. Cambridge, MA: MIT Press.

Shuwairi, S., M. Albert, and S. Johnson. 2007. Discrimination of possible and impossible objects
in infancy. Psychological Science 18 (4): 303–307.

von der Heydt, R., F. Qui, and K. Endo. 2001. Depth perception with ambiguous displays in
humans and monkeys. Perception 30 (ECVP abstract supplement): 41.
9 Observational Concepts

Daniel A. Weiskopf

9.1 From Perception to Thought

How is it that we are able to think about what we perceive? More specifically, how
are we able to bring the resources of conceptualized thought to bear on the objects
and events that are represented to us in perception? And how much of our capacity
for conceptualized thought is undergirded by, or is an extension of, our capacities for
perception and action? Addressing these questions requires disentangling some of the
more tightly woven strands linking perception, thought, and action.
Although one of the most distinctive features of human concepts is that they extend
over indefinitely many types of things that transcend perception, we can also reflect
on straightforwardly perceivable objects, and the concepts we have of these objects
are often acquired by perceptually encountering and manipulating them. Additionally,
how we perceive the world is infused or colored by the conceptual capacities that we
possess. We don’t merely see the cat, we see her as a cat, a visual state that depends on
conceptualizing her in a certain way. And in virtue of seeing her as a cat, we may come
to form perceptual beliefs concerning her presence and qualities. Thus, conceptualized
perception enables conceptualized thought.
My aim here is to illuminate what happens at the interface between perception
and higher cognition. On the view I develop, observational concepts are the pivot on
which this relationship turns. Observational concepts are those that are spontaneously
made available at the interface between perception-action systems and the conceptual
system. They correspond to the ways we have of conceptually dividing the world based
solely on the available perceptual information. We are able to treat what is perceived as
evidence for what isn’t directly perceived (or, indeed, perceivable at all). Observational
concepts form the evidential basis for these perception-transcendent inferences. And
ultimately, we acquire ways of thinking about perceived things that are not tied directly
to how they are perceived; observational concepts are central to this process insofar
as they provide us with an initial way of conceptually tracking categories that we will
learn to represent in perception-transcendent ways.
224 Daniel A. Weiskopf

In what follows, I situate observational concepts in the larger architecture of

cognition, characterize their role and content, describe how they are learned, and show
how they play a key role in learning further concepts. Along the way I distinguish
them from related constructs such as recognitional concepts and show that arguments
against recognitional concepts fail to work against observational concepts. I conclude
by discussing what observational concepts can teach us about the extent to which
perceptual systems and representations may shape higher cognition.

9.2 Interfaces

Observational concepts are distinguished by their functional role, specifically by the

location they occupy in the overall architecture of cognition. Many issues about cog-
nitive architecture remain largely unsettled, but the only architectural assumption
employed here is the distinction between input-output systems and central cognitive
systems. Central systems include but need not be exhausted by the conceptual system,
which is not assumed to be unitary.1
In the simplest case, the mapping from perceptual systems to central systems is
direct, so that the hierarchy of levels of sensory processing culminates in contact
between sensory systems and systems of conceptual interpretation. There may, how-
ever, be complex perception-action connections that bypass higher thought, as in
some forms of online sensorimotor coordination. On a more indirect arrangement,
perceptual systems feed first into intermediate nonconceptual systems that preprocess
sensory inputs by transforming, modifying, tagging, and re-representing them in vari-
ous ways.2 Whether the arrangement is direct or indirect, however, there must be some
sort of interface between the conceptual system and these various nonconceptual pro-
cessing systems. What goes on at such an interface is precisely a transition from the

1. The main views ruled out by this assumption are subsumption-style architectures and varia-
tions on them, that is, any model on which there are mainly layers of horizontal connections
running directly from sensory input to motor output, with few connections between these layers
(Hurley 2008). Although I will mostly speak of the conceptual system here, massive modularists
may replace this with talk about a host of central modules.
2. This is one way of viewing the systems of core cognition discussed by Carey (2009). Core
cognitive systems are domain-specific mechanisms that analyze perceptual inputs and produce
conceptual outputs, including foundational concepts of objects, number, and agency. While
Carey holds that these systems have “integrated conceptual content” (2009, 67), she also holds
that they use iconic but nonperceptual representations. Architecturally, they are an intermediate
processing stage between perception and central cognition.
Observational Concepts 225

way the world is represented in perception (or nonconceptually) to the way that we
conceive of the world.3
The outputs of perception, while nonconceptual, are enormously rich, presenting
us with a three-dimensional world that comes divided into distinct objects and events
that are assigned locations and motion vectors in space, along with various perceivable
qualities: color, shape, shading, texture, degree of solidity, and so forth (Clark 2004;
Matthen 2005). These outputs constitute interface representations, inasmuch as they
are accessible to both conceptual and nonconceptual systems. The job of these repre-
sentations is to present a limited body of information from one system to another in a
legible format, where a representation’s format is the way in which its semantic content
is encoded in the structure of the vehicle. The way a system formats information is
tailored to the problem it solves or the task it carries out. Candidate formats for mental
representations include icons and images, propositions, frames, maps, models, analog
accumulators, and so on.
Interfaces can be classified in several ways. First, systems can make use of the same
representational vehicles or different ones. Consider a parsing mechanism that con-
tains syntactic and semantic components. The syntactic parser might construct a tree
that represents constituency, dominance, and other formal relations, which is then
passed to the semantic parser for interpretation. Here the same representation is trans-
ferred from one system to another. Among interfaces that use the same vehicles, some
involve passing along all of their representational outputs across the interface, while
others involve passing only some of them. The latter case can arise where one system
interfaces with two downstream systems, each of which has access to only some of its
outputs. This occurs in vision, where after the common retinocortical pathway and
processing in V1, the visual output divides into the dorsal and ventral streams, which
make use of different properties of the visual input (Milner and Goodale 2006). Where
an interface involves passing all of one system’s output to another, making use of the
same vehicles, call this a pass-through interface; where only a subset of these repre-
sentations are disseminated, call this a filter. Systems that compute intermediate level
representations often impose filters that prevent these from being passed downstream,
and some systems filter their outputs differently depending on the consumer systems
they are feeding.

3. I have been framing things here in a way that takes sides in the debate between conceptualists
and nonconceptualists about perception. While there are many ways of drawing this distinction,
for present purposes we can say that nonconceptualists hold that perceptual states are not indi-
viduated by reference to the conceptual repertoire of the cognizer, while conceptualists hold that
they are. I am assuming nonconceptualism in the following sense: there is at least some range of
perceptual states available to cognizers that is independent of the conceptual capacities that they
possess; being in these states does not involve deploying any concepts.
226 Daniel A. Weiskopf

Finally, some interfaces occur between systems that make use of different
representational vehicles. Call these translational interfaces: ones that transform infor-
mation encoded in one format to information encoded in another. The simplest exam-
ple is a digitizer, which converts a continuous analog signal into discrete bits. Cognitive
systems employ many different encoding schemes for different kinds of information,
so there must exist at least some translational interfaces of varying degrees of com-
plexity. In numerical cognition, there are mappings from a display of a quantity to an
analog accumulator; in visually guided reaching, a perception of an object in space is
mapped to a motor representation of how it can be grasped; and in language produc-
tion, a semantic representation is mapped to a phonological word form. These systems
involve different codes, so their interfaces are translational.
This way of thinking about interfaces has a nice virtue: it just happens to correspond
well with the major historical proposals about how concepts are acquired from percep-
tion. Empiricist theories of concepts adhere to the shared vehicles thesis: conceptual
thought reuses the representational vehicles deployed in perception (Barsalou 1999;
Prinz 2002). On classical empiricist views such as Locke’s and Hume’s, the relation-
ship between percepts and concepts is described in several ways, the simplest of which
involves the process of copying. In Hume’s terms, ideas are less vivid copies of impres-
sions, where copying preserves an impression’s form but drains some of its “force and
vivacity.” Concepts are copies of percepts that play a distinctive functional role in cog-
nition, and since copying preserves most of the properties of representational vehicles,
acquiring a concept from experience involves a pass-through mechanism that causally
reproduces those vehicles in a different system.
Beyond simple Humean copying, Locke sometimes describes the acquisition of
complex ideas as a kind of abstraction from experience with particular category
members. A Lockean abstraction mechanism selectively deletes distinctive features and
retains common ones, and in this sense it is a filter for these features. The output of this
process is a general concept: a stripped-down perceptual representation that captures
the characteristics shared by the instances and other category members. In much this
way, Locke describes the learning of a general idea like HUMAN as a process of compar-
ing ideas of individual persons (Mama, nurse, etc.) and subtracting their distinctive
qualities. As an approximation, then, we can think of Humean copying as a pass-
through mechanism, and Lockean abstraction as a filter.4
Nonempiricist views of concepts, by contrast, see this interface as being transla-
tional, since they hold that thought employs an amodal code distinct from those used
in perception. So Fodor (2008) distinguishes between perceptual and conceptual rep-
resentations on the grounds that the former are iconic while the latter are discursive.

4. This is a simplification in many ways, not least of which that Locke seems to have had several
theories of conceptualization running at once. See Gauker (2011, chapter 1) for an excellent
Observational Concepts 227

In iconic representations, every part of the representation stands for a part of what is
represented, while discursive representations have a “canonical decomposition” into
constituents, which are the only parts of the representation that contribute to its over-
all semantic interpretation. Photographs are iconic: in a photo of a koala, every part
of the photo represents some part of the koala; the part containing its round, fuzzy
ears represents those very ears, their shape and texture, and so forth. Sentences, on the
other hand, are discursive, so that in the sentence “the koala has round and fuzzy ears,”
the string the koala represents a contextually specified bear, but the string has round and
represents nothing, since it is not a semantically relevant part. Although every part of an
icon is semantically significant, not every part of a discursive representation is. Accord-
ingly, in moving from perceptual/iconic representations to conceptual/discursive ones,
a semantically homogeneous input is discretized by a translation mechanism.5
The minimal conception of an interface is a device that mediates information
transfer and control between two or more systems by letting the output states of the
producer systems determine the input states of the consumer systems. An interface is
a normal route for this kind of configuration. And the taxonomy of possible interfaces
is one that maps onto the historically dominant accounts of conceptualization in the
empiricist and rationalist traditions.

9.3 Observational Concepts Introduced

We can now reintroduce the idea of an observational concept. Where perceptual

systems make contact with central cognition, there is an interface whose function is to
generate a conceptual output in response to that type of perceptual input. Given the
structure of the interface and the right background conditions, certain percepts will be
causally sufficient to produce conceptual output states in a way that is not mediated by
any other concepts.6 Observational concepts are those concepts that are spontaneously
activated or made available for use solely on the basis of a subject’s being in a certain
perceptual state.
Such concepts are the basic constituents of perceptual judgments, such as the belief
that there is a cat on the green chair by the window, that those are sunflowers, or that
there was just a loud bang to our left. These are our perceptually informed opinions

5. One needn’t hold that concepts are encoded discursively to think of this interface as transla-
tional. Mental model-style views of higher cognition might employ map-like representations that
nevertheless differ from the iconic representations employed in perception. See also Mandler
6. Any such mediation, should it exist, cannot go on forever—if one concept’s activation is
itself conceptually mediated, then that mediating concept is observational. Given the nature of
interfaces, we can know that this regress must terminate, since there cannot be infinitely many
representations lying along this causal pathway.
228 Daniel A. Weiskopf

concerning the objects and events that surround us; they are the judgments that we
can be in a position to make about the environment just on the basis of what percep-
tion plus our repertoire of observational concepts make available.7
There are two counterfactual aspects to this specification of observational concepts.
First, these concepts are not invariably tokened when the right perceptual inputs are
present. Perceiving a tree does not always lead to tokening TREE, still less to thinking
THAT IS A TREE. The reference to background conditions becomes important. These con-

ditions have to do with resources like attention and memory as well as motivational
factors, goals, interests, and other ongoing thoughts one is entertaining. Perceptual
inputs are sufficient for activating the appropriate observational concepts as long as the
background conditions are those that dispose the creature to be conceptually respon-
sive to its perceptual input. This is compatible with the possibility that perceptual states
only ever actually lead to thoughts involving such concepts against a background of
already active thoughts, plans, and broader aspects of one’s mental “set.” It only needs
to be possible for these concepts to be entertained in isolation from this surrounding
mental activity.
Perceptions don’t dictate what perceptual judgments follow from them, nor how the
perceived scene will ultimately be conceptualized. A trivial example: the visual percep-
tion of the cat dozing on the chair may give rise to the belief that the cat is on top of
the chair or the (distinct) belief that the chair is under the cat. The cat itself might be
represented as a cat, as an animal, or as Sophie. The same auditory perception might
be equally well conceptualized as a sigh or a snort. All that is required is that some
concepts are immediately made available by the occurrence of a perceptual state,
though a representation’s being made available, in the present sense, doesn’t entail its
being deployed for any particular purpose.
Second, perceptual input itself is not necessary for the activation of an observational
concept, since being a concept requires availability for many other cognitive processes,
including offline reasoning and planning, direction of other actions such as the forma-
tion of mental images, and so on. These tasks can take place under endogenous control,
in the absence of any ongoing perception. So observational concepts have a distinctive
set of application conditions that tie them to perception, but they also enjoy the free
deployability characteristic of concepts generally.
Observational concepts are not merely occasioned by perceptual experience but are
in a stronger sense directed at or applied to what is perceived. In a weak sense, a per-
cept can be associatively linked with any sort of mental or behavioral state. There are
direct behavioral associations such as the photic sneeze reflex, or potentially arbitrary

7. For an excellent discussion of the epistemology of perceptually basic beliefs in this sense, see
Lyons (2008). The view being developed here does not purport to have any special epistemologi-
cal status or import, although it might be developed in those directions.
Observational Concepts 229

psychological associations such as thoughts of springtime coming to mind whenever I

see lilies. To forge a link stronger than mere association between perceiving an object
and forming perceptual judgments about it, there needs to be a semantic relationship
of co-application or co-indexing. A percept and concept are co-indexed when the con-
cept is, or is disposed to be, applied to the very same object that the percept is tracking.8
A function of interfaces is to provide such semantic relationships so that information
about the same object can be collected across different cognitive systems and represen-
tational formats. When vision or any other object-centered sense divides the perceived
environment into an array of entities, they are passed on to the conceptual system
along with individuating tags that distinguish them as separate possible objects of clas-
sification and predication.
Finally, this division of representational labor gives us an account of recognitional
seeing, or seeing as. To see a tree, and to see it as a tree, are two different states. The latter
is more complex in that it contains the former. Seeing x involves having a representa-
tion of the perceptual qualities of x that is indexed to the object x itself; seeing x as F
involves having this percept co-indexed with the concept F. States of seeing as are thus
perceptual-conceptual hybrids, having one foot in each system. But notice that seeing x
as F falls slightly short of seeing that x is F. Seeing as is a (partially) conceptualized state,
but not a propositional state. Activating a concept, making it ready for use, is not the
same as employing it in a judgment, which is the form of seeing that. In this sense, you
can recognize x to be F without entertaining the proposition that x is F.

9.4 Against Perceptual Inflation

Observational concepts are limited by the perceptual similarities we can reliably track.
In particular, it would seem we cannot have observational concepts of abstract entities
(those that have no physical, perceivable manifestations at all) and categories that have
overly heterogeneous perceptual presentations. Call this, following Jesse Prinz (2005),
the imperceptibility thesis. Prinz denies the thesis and argues that we can perceive indefi-
nitely many abstract categories; indeed, virtually anything we can form a concept of

8. Much work in midlevel vision has emphasized the need for such co-indexing representations
(Pylyshyn 2001; Scholl 2001). This work is obviously relevant here, with the caveat that the
instantiation tags referred to in this literature are usually opened when objects move, change, or
otherwise become salient. I would expand this notion so that any object that can be distinguished
from its environment can be tagged and be the subject of conceptual predication. In fact,
co-indexing relationships are needed for two cognitive purposes. One is to ensure that multiple
representations track the same object across processing systems and contexts. The other, empha-
sized by Clark (2004), is to ensure that separate representations are linked synchronically into a
single representation of an object; that is, to solve the binding problem.
230 Daniel A. Weiskopf

can be perceived. This includes abstract categories of things such as numbers, or prop-
erties such as being a philosophy major or being an uncle.
Prinz’s argument turns on his view about recognitional perceiving. One perceives X
just in case: (1) X impinges on one’s sensory transducers, (2) leading to the formation
of a perceptual representation, (3) which is in turn matched against stored representa-
tions of X. So to perceive something involves forming a percept of it and retrieving a
representation of it, where the content of the retrieved representation determines the
content of the perception one has. In addition, Prinz follows Dretske’s (1986) informa-
tional semantics, on which M represents C just in case instances of C have the power to
cause instances of M, and this relationship holds because M has the function of detect-
ing C. Content is a matter of teleologically established and sustained informational
Many abstract and intuitively imperceptible qualities will turn out to be perceivable
on this view, since there is always some perceptual scenario that can cause us to retrieve
representations of them. Consider being an uncle. Uncles have no common perceptual
look, so I cannot just pick them out in a crowded room. However, if I arrange for all
and only the uncles to raise their hands (assuming everyone is sincere and coopera-
tive), they suddenly share such a look. If I arrange to get myself to think someone is
an uncle just in case their hand is raised, and hand raising is a detectable look, then
what I am perceiving is unclehood, since my UNCLE concept is activated when I see their
raised hands.
Now consider number. Small numbers might be perceivable; triples of discrete
objects have a distinctive look, so perhaps we can perceive threeness and form the
corresponding observational concept. Few think that we can perceive or form such
concepts of higher numbers such as fifty-seven. But we can arrange a situation in which
we will activate FIFTY-SEVEN in response to a perceptual scenario: we simply count the
objects, keeping track of them using number words. Our saying the numerals out loud
or internally is a perceptual state, so when we run out of things to count, we have
perceived the number corresponding to the last numeral that was perceptually acti-
vated. Generalizing these examples shows that we are able to perceive both abstract
entities and qualities, so the imperceptibility thesis would be false.
But Prinz’s account of perceptual content is implausibly permissive. Neither percep-
tion nor observational concepts themselves have contents as rich as the arbitrarily
abstract categories that we can conceive of. The problem lies with the open-ended
notion of recognitional perceiving. Recall that any sort of representation that can be
retrieved as part of the comparison or retrieval process initiated by perception can
contribute to what is perceived, and there are no limits on how complicated these
processes may be. This lack of limits is what lets us perceive uncles by seeing raised
hands—but since this connection depends on a complex cognitive arrangement that
goes beyond what the perceptual system itself contributes, we should not consider it
Observational Concepts 231

an act of perception, properly speaking. Similarly in the number case: the act of count-
ing large numbers uses more cognitive resources (memory, language, etc.) than just
what our perceptual systems provide.9 Worries about this form of content inflation are
blocked on the account of observational concepts given here by the requirement that
their deployment be under the control of an interface, rather than allowing any down-
stream representation to count.
Susanna Siegel (2010) has also argued that the contents of perception are extremely
rich, so rich that we can perceptually represent not just colors, shapes, textures, and
the rest of the qualities that traditional theories of vision admit, but also so-called
K-properties. These include properties covering objects (natural and artifact kinds),
actions and events, mental states, and semantic facts, for example: person, mountain,
bicycle, carrying a dog, being inquisitive, being a word of Russian, being a phrase meaning
“the highway exit is ahead.” According to the rich content view, all of these are possible
contents that we can grasp in experience, and we can recognize this fact from attend-
ing to the phenomenology of the relevant experiences.
Siegel’s argument runs as follows. Consider two related experiences: in E1, you are
looking at a bunch of pine trees, but you do not have the ability to visually recognize
pine trees; and in E2, you are looking at the same trees, but you have learned to recog-
nize them. There is, it would seem, a phenomenological difference between these two
experiences. Being able to perceptually recognize what you are looking at is different
from not knowing what you are looking at, and the overall phenomenological state
one occupies in each case seems different. To arrive at the rich content view, three
further premises are needed: (1) if E1 and E2 differ in phenomenology, then there is
a phenomenological difference between the experiences themselves; (2) if there is a
phenomenological difference between them, then they differ in content; and (3) if
they differ in content, it is a difference with respect to the K-properties that those expe-
riences represent. Since we have granted that E1 and E2 differ in phenomenology, we
quickly reach the conclusion that experience represents K-properties.
The account of how concepts come into play in acts of perceptual recognition gives
us a response to this argument, one that focuses on the denial of premise (1). We
should grant that there is a phenomenological difference for the perceiver having E1

9. Prinz does argue separately that our sensory systems proper can represent these abstract
categories as well. Here he relies on the existence of downward projections from higher cortical
centers to the sensory systems themselves. These downward projections allegedly allow concep-
tual information to “infuse” sensory processing. However, on Prinz’s official theory of content, it
is hard to see how this might work. If the dominant cause of activity in sensory systems is the
perceivable object, then that is what sensory representations refer to, not abstractions. If their
cause is split among external objects and higher-level systems, the theory does not assign them
content, since their function and informational role is indeterminate.
232 Daniel A. Weiskopf

versus E2, but deny that this difference is located in the visual experience proper.10
Siegel, in sketching possible responses, considers two related objections to this premise.
Both involve saying that the phenomenological difference between E1 and E2 lies not
in the visual experience itself but in some associated psychological aspect, in particular
in some sort of cognitive experience. If E1 and E2 differ only in this related respect, there
is no difference in visual content, particularly not with respect to the representation of
The candidate cognitive experiences Siegel considers, however, are all forms of
propositional representation. They are either commitment-involving attitudes (e.g.,
beliefs or judgments) or noncommittal attitudes (e.g., hunches or merely entertained
thoughts). Against the idea that the phenomenological difference comes from the pres-
ence of a commitment-involving attitude such as the belief that that is a pine tree,
notice that I might well believe (because I am reliably informed) that the tree appear-
ances I perceive are not real trees, but only props or holograms. Despite this belief,
they might appear phenomenologically different in situations E1 and E2. So committal
attitudes such as beliefs cannot make the difference between the two.
Against the idea that noncommittal attitudes might explain the difference, she
presumes these attitudes to be occurrent thoughts such as that kind of tree is familiar.
Thoughts of seeming familiarity say nothing about what the perceived objects them-
selves are. However, she argues, these thoughts are simply unnecessary: “There need
not be, it seems, an extra episode (or occurrent state), beyond sensing, for the phenom-
enological change to take effect” (106). So any alleged noncommittal attitude would be
redundant in explaining the E1/E2 difference. Given that (with a few caveats) these two
possibilities exhaust the primary forms of nonsensory experience, we can conclude that
the phenomenological differences are located in the visual experiences themselves, not
in any adjoining cognitive states.
Observational concepts have a distinctive functional role that renders them well
poised to thread this needle, however. First, as noted above, they are not propositional
representations, and they do not inherently carry committal force. They may do so, if a
concept becomes activated to the point where it is actually endorsed and applied to a
perceived scene. The application of a concept to perception is a state that has distinc-
tive correctness conditions, and thinkers are committed to these. But since activation
is graded, these concepts may be exerting an effect even though they are not past the
threshold for being applied. In the hologram case, once I am told that the trees are
fake, I refrain from applying the concept, but it may retain a residual level of activation.

10. I should add that talk of experiences here is Siegel’s; I am glossing visual experiences as being
states that depend for their instantiation on activity in visual systems; similarly, cognitive experi-
ences are states that depend on nonperceptual systems in some way. This gives a way of mapping
experience talk onto talk of underlying cognitive systems.
Observational Concepts 233

What, then, of the objection that these concepts constitute merely a redundant
noncommittal component? It is not clear that we have a separate occurrent state
here at all. Observational concepts are distinct from their perceptual antecedents,
true, but their activation levels are gradual. So the simple notion of a state’s being
(non)occurrent does not obviously have application here. Certainly the issue cannot be
settled by intuitions about how many “episodes” are involved in an experience. Siegel
suggests that if a nonsensory event is not explicit and occurrent, then “it becomes less
clear that it is phenomenally conscious at all” (107). But why is this? The phenom-
enal properties of cognitive states, or their contribution to overall phenomenology,
may in principle depend on any aspect of their functional or representational role. To
take an example, tip-of-the-tongue states are cognitive, but they have a highly distinc-
tive phenomenology that is plausibly underpinned not by explicitly entertaining any
proposition, but rather by cascades of activation washing through networks of lexical
and semantic memory. If these types of spreading activation can contribute to cogni-
tive phenomenology, the same should be true of the varying levels of activation in the
case of observational concepts.
What these two cases show is that attending to the role of observational concepts
can help to establish boundaries on the representational power of perception. Architec-
tural divisions determine what can be recognitionally perceived, thus blocking the sort
of content inflation Prinz’s account invites. Similarly, a wider range of functional
interactions between perception and cognition can help us to account for the phenom-
enology of perceptual recognition without endorsing Siegel’s rich content view.11 The
phenomenological evidence that Siegel draws our attention to needs to be explained,
but only by uncovering the details of the larger architecture and how it assigns content
to underlying systems and states.

9.5 Observational versus Recognitional Concepts

Observational concepts in my sense are not the same as recognitional concepts. Fodor
presents a view of recognitional concepts according to which “a concept is recogni-
tional iff: (1) it is at least partially constituted by its possession conditions, and (2)
among its possession conditions is the ability to recognize at least some things that fall
under the concept as things that fall under the concept” (Fodor 1998, 1). Recognitional
concepts are those that are in part constituted by abilities to recognize some things in
their extension. If RED and SQUARE were recognitional, then possessing them would mean

11. Although I am not necessarily endorsing the idea that all of Siegel’s K-properties can
be the content of observational concepts, in my sense. Addressing this issue would require
more clarity both on the scope of observational concepts, and on how to fill in the list of
234 Daniel A. Weiskopf

necessarily being able to apply them to red things and squares, respectively, on the
basis of the appropriate perceptual encounter.
Fodor argues that there cannot be any such recognitional concepts, on the grounds
that recognitional abilities fail to be compositional. Since compositionality is a neces-
sary condition on concepts, anything noncompositional cannot be part of what indi-
viduates concepts. The main premise of his argument is that the possession conditions
for a complex concept must include the possession conditions for its constituents. So
whatever states, capacities, dispositions, and so forth are required for having complex
concepts such as RED SQUARE, FISH FOOD, or ONE-ARMED MAN necessarily include whatever is
required to have the concepts that make them up. Possession conditions are inherited
compositionally. But the ability to recognize instances is not inherited in this way. One
can be able to separately recognize fish and food but not be able to recognize that those
confetti-like flakes are fish food. Recognizing fish food depends on knowing something
about what fish actually eat, not on anything one could extract from the constituent
concepts themselves. Since anything that is a constitutive property of a concept must
be inherited compositionally by its hosts, recognitional abilities cannot be concept
This argument doesn’t work, however.12 Suppose that some concepts have recogni-
tional capacities c1–cn among their essential properties, and that these capacities must
be among the possession conditions for any complex concept that hosts them. Suppose
too that there is no corresponding recognitional capacity c* for the complex concept
itself—or if there is one, it is neither among c1–cn, nor is it derivable from them alone.
This shows that not every concept composed from recognitional concepts is itself
thereby recognitional. Even so, recognitional capacities are “inherited” in the relevant
sense—having a complex concept entails having the capacities that are the possession
conditions for its constituents, and this helps to explain why having the constituents
is roughly sufficient for having the complex concept. So the fact that a wholly distinct
recognitional capacity for fish food fails to be produced compositionally from the com-
bination of FISH with FOOD doesn’t argue against the existence of recognitional concepts,
so long as having FISH FOOD entails having the appropriate recognitional capacities for
fish and food, taken individually.13

12. The argument given here is similar to one advanced by Recanati (2002). For careful analysis
and criticism of the notion of a recognitional concept, see Dahlgrun (2011); for a Fodorian
defense of one type of recognitional concept, see Rives (2010).
13. Fodor does in fact think that “people who have the concept PET FISH do generally have a cor-
responding recognitional capacity” (1998, 8). It just can’t be one they’ve derived solely from the
recognitional capacities of that concept’s constituents. But this is only a problem if one wants to
maintain that all complex concepts formed from recognitional concepts are also recognitional,
which there is no reason to do. Fodor argues that this hybrid view is arbitrary or theoretically
inelegant, but the architectural considerations pointed to above show that it is, in fact, quite
Observational Concepts 235

The model of observational concepts gives a principled explanation for why we

would not expect recognitional abilities to compose in Fodor’s sense. An observa-
tional concept for FISH allows one to categorize things as fish, as long as there are good
instances presented in good perceptual conditions. This ability is mediated by the
structure of the interface, however, which takes a restricted set of percepts as input.
Imagine the interface as consisting of a set of perceptual analyzers attuned to category
appearances. These perceptual analyzers may respond to what counts as good instances
of fish, and so similarly would analyzers that take appearances of food as input. But the
existence of these two analyzers entails nothing about the existence of a third device
for responding to the characteristic appearance of fish food; in fact, given that it looks
rather unfoodlike, one would predict that there isn’t any such device, and hence that
there is no observational concept FISH FOOD.
In any event, however, observational concepts are unlike Fodorian recognitional
concepts, and also unlike Peacocke’s (1992, chapter. 3; 2001) perceptual concepts
in that their identity as concepts is not constituted by their perceptual connections.
Observational concepts have a certain primary perceptual affiliation, namely, a means
of being directly deployed by perceptual inputs. This affiliation makes them observa-
tional, but it is not part of their possession conditions qua concepts.
Rives (2010) has defended a Fodorian notion of recognitional concepts. He argues
that causal links to perception are essential to some concepts, on the basis of the
general principle that scientific kinds are individuated by the laws that they participate
in. For concepts, these laws include those that fix their content and causal role by
connecting them with particular perceptual states. So principles of kind individuation
mandate that some concepts are recognitional. Undoubtedly, the general point that
kinds in the special sciences are taxonomized by their causal and functional profile is
correct. But many of these causal links are fragile, and we should be wary of making
concept identity contingent on them.
For an actual case in which observational concepts have their perceptual affilia-
tions severed, consider associative agnosia. The core disorder involves the inability
to identify visually presented objects either by naming or by nonverbal means (e.g.,
sorting). Patients tend to perform best with real objects or highly realistic ones such as
photographs; when they hazard an identification, they often confuse visually similar
objects. Associative agnosics can sometimes describe the features that objects have but
seem not to know what the described objects are. The disorder is multifaceted, and its
basis is not entirely clear (Farah 2004, chapter 5; Humphreys and Riddoch 1987, 2006),
but its existence indicates that specifically visual routes to conceptual deployment can
be disrupted without corresponding loss of the concepts themselves.
The patient CK, for example, was capable of providing elaborate verbal descriptions
of objects that he could recognize by touch but not by sight. He could also answer
specific questions about the visual properties of objects, suggesting that this
fine-grained visual knowledge was usable in reasoning tasks but not in object
236 Daniel A. Weiskopf

identification (Behrmann, Moscovitch, and Winocur 1994). The patient ELM, who
played in a military brass band, was able to freely describe the nonvisual properties
of brass instruments in detail, and to use concepts of these instruments in an asso-
ciative learning task despite his deficits in visual identification (Dixon et al. 2002).
The converse pattern also appears: in some cases of category-specific semantic impair-
ments, there is loss of almost all information that might be used in reasoning about
a category, despite relatively spared ability to identify category members (De Renzi
and Lucchelli 1994). So the capacity to identify category members and the capacity to
reason about them are at least partially separable.
These cases suggest that agnosic patients have not lost the concepts of the objects
that they can no longer identify, and that concepts acquired observationally can persist
once these core functional links are severed. Indeed, it seems to be a general truth that
concepts are not particularly fragile; they can survive arbitrarily many gains and losses
in their representational affiliations. As a design feature, this makes sense: the more
such essential links a concept has, the more difficult it is to acquire and the easier it is
to lose. Concept possession should be buffered from such changes in cognitive struc-
ture, however. Given this, the notion of an observational concept seems to be a marked
improvement on both the notion of a recognitional concept (as Fodor conceives it) and
of a perceptual concept (as Peacocke conceives it). Both of these notions commit us to
constitutive links between concepts and perception, but these, I have argued, we have
ample reason to reject.

9.6 Learning Observational Concepts

Observational concepts link the conceptual system to the world in a flexible, open-
ended way. Learning an observational concept involves two stages: first, the construc-
tion of a dedicated perceptual analyzer that can co-classify objects based on their
appearances; second, the construction of a link between the output of that analyzer
and a symbol in the conceptual system.
The main job of a perceptual analyzer is to create structure in a perceptual
similarity space. There are many mechanisms for achieving this. Local, bottom-up pro-
cessing has been most intensely studied in theories of object recognition, where the
debate has focused on whether multiple stored views of an object or single viewpoint-
independent structural descriptions are required (Feldman 2003; Riesenhuber and
Poggio 2000; Pessig and Tarr 2007; Tarr 2003). These representations also need to be
processed by the right sort of generalization mechanism or similarity gradient, along
with a weighting matrix for assigning parts of the object greater or lesser significance.
Once a set of views of an object have been stored and linked together, or once a struc-
tural description has been generated, the process of analyzing new percepts using this
device depends just on the fit between the percept and the stored representations.
Observational Concepts 237

However, perceptual analyzers may be constructed from many different materials.

How an object fits into a larger perceived scene is also relevant. Objects that appear in
a familiar location, or in the context of other familiar objects, tend to be recognized
faster and more accurately (Henderson and Hollingworth 1999). Analyzers take into
account holistic scene-level properties in determining object categorization, not just
local qualities. More interestingly, object recognition also depends on whether some-
thing is appropriately poised for action (Humphreys and Riddoch 2007). Recognition
can be facilitated by the orientation of an object relative to the normal actions one
performs with it, and by its orientation relative to other objects in a perceived scene,
even if they are not the usual ones that it is seen to interact with.14
Perceptual analyzers, then, may have extremely rapid access to a large amount of
information. They can integrate stored viewpoints and structural descriptions, as well
as holistic scene-level properties and associations and stored or potential patterns of
motor interaction and event schemas that objects can enter into. Neurally, the imple-
mentation of visual analysis seems to involve such widespread retrieval insofar as
it taps massively sweeping feed-forward projections as well as recurrent projections
that play a role in mediating visual awareness (Lamme 2006; Lamme and Roelfsema
2000). These analyzers may be assembled relatively quickly—seven hours of training is
sufficient to induce new category-specific responses in the fusiform gyrus (Gauthier
et al. 1999). None of this processing, however, needs to be conceptually mediated.
Extraction of viewpoints and structural descriptions can occur prior to knowing what
the object being analyzed is, as can scene analysis and co-occurrence detection. Action-
based understanding may stem from keeping track of our own past actions with
perceived objects as well as from on-line computation of the plausible acts and events
that objects afford.
The traditional view about the scope of observational concepts is that they are
situated at the “basic level” (Jacob and Jeannerod 2003, 140–141; Rosch et al. 1976).
Basic categories of objects such as cats, birds, hammers, airplanes, and squares can be
grouped together based on the similarities and differences detectable by the percep-
tual system itself, even at three to four months old (Eimas and Quinn 1994; Quinn,
Eimas, and Rosenkrantz 1993). But narrower object categories also involve such simi-
larities: ball peen hammers versus claw hammers, MD-80s versus Boeing 747s, domestic
shorthairs versus Manx cats. Expert object classification treats such specific groupings

14. I should add that the view of perception I am adopting here is compatible with the existence
of many tight links between perception and action systems. Indeed, there is robust evidence that
the two are tightly coupled, so that sensorimotor integration rather than separation might be the
norm (Prinz 1997; Schütz-Bosbach and Prinz 2007). Observational concepts, nevertheless, occur
where perceptual systems interface with concepts, no matter how richly connected perception
and action themselves may be.
238 Daniel A. Weiskopf

as entry points. Most of us behave like experts when it comes to classifying well-
known individuals, as shown by the fact that we can rapidly and correctly identify
Bill Clinton, Golda Meir, the Mona Lisa, or the Sears Tower across different encounters
(Bukach, Gauthier, and Tarr 2006; Tanaka 2001; Tanaka and Gauthier 1997. Even
relatively global perceptual similarities may be detected early, such as those that unite
all entities that move in the same way as animate living beings, or those that move in a
way distinctive of vehicles (Behl-Chadha 1996; Mandler and McDonough 1993, 1998).
Observational concepts naturally coalesce around such islands of perceptual similarity.
Having a set of perceptual analyzers causes perceived objects to cluster in similar-
ity space, but it does not give one the ability to think about those objects until they
are assigned to conceptual representations (Gauker 2011, chapter 3). This requires
noticing the similarities that are being detected by a perceptual analyzer and coining a
new mental symbol for the purpose of keeping track of that kind of thing, namely, the
kind of thing that has the relevant appearance that the underlying analyzer tracks.
Conceptualization thus involves both a functional transformation and a shift in
content. Perception knows about looks: it informs us about how things appear, and it
tracks and produces detectable similarities among appearances. Concepts, in the first
instance, are not for tracking appearances, but for determining what things are, and
for representing them in a way that is autonomous from perception. So the content of
perception informs us about the appearances of objects around us, but when we coin
observational concepts they must be (as the agnosia cases show) potentially detachable
from these appearances: they track categories of things that have, but are not defined
by, those appearances. Perception groups things that look F-like, while concepts track
things that are Fs. Part of what is involved in generating observational concepts from
perceptual analyzers is shifting from content that represents only looks to content that
represents categories as distinct from their looks.15
This point can be illustrated using Peacocke’s account of perceptual content (1989,
1992). For Peacocke, the accuracy conditions for perceptual states are given in terms
of abstract scenario types. A scenario is a way of filling out space in the vicinity of the
perceiver; it specifies a way of locating surfaces and their properties relative to an origin
point and a set of orienting axes. A perceptual scene, then, is a way that this spatial

15. One may object here that some perceptual representations seem to track more than mere
appearances. Some models of object vision propose that there is a stage at which object-centered
visual representations such as 3D models of object parts and structure are computed. These are
independent of the particular viewpoint of an observer and hence may seem to be about more
than looks. The difference, however, is that even these 3D models do not distinguish between
being a member of one category that has a certain 3D profile and being a member of a different
category that just happens to share that profile. That is, these 3D models are not yet kind repre-
sentations and hence are still about appearances in the generalized sense meant here. Thanks to
Jake Beck for raising this objection.
Observational Concepts 239

volume may be filled in. A filled-in scene depicts the world as containing surfaces at
various distances and orientations, instantiating properties such as color, texture, shad-
ing, and solidity. This captures the look of the scene, but it does not specify what kinds
of objects the world itself contains. The very same perceptual scene is compatible with
many different conceptualizations. This is true not just in the sense that there may be
various different entry points into scene categorization, but also in the sense that how-
ever things happen to look in a scene, they may not be what they appear on their sur-
faces to be. The cat may be a puppet, the tree a prop. Our recognition of this face-value
gap is possible because the function of concepts is to provide possibilities for represen-
tation and control of thought and behavior that go beyond what perception delivers.
Conceptual content, then, reflects the appearance-transcending role of the con-
ceptual system. This point has nothing to do with the standard way of drawing the
distinction between perceptual and nonconceptual states that relies on appeals to the
fine-grainedness or richness of perceptual representation (Tye 2005). There is no inher-
ent reason that rich, fine-grained properties cannot enter into the content of some of
our concepts. This is suggested by the fact that observational concepts present their
content under a perceptual mode of presentation. Modes of presentation generally are
ways in which categories are grasped in thought; they are how we are thinking about
a particular category. An observational concept presents its content as being the kind
of thing that looks like THAT, where THAT refers to the complex representational con-
tent that is computed by the relevant perceptual analyzer. Hence, having the concept
normally involves having access to facts about looks. Given this access, we are also
capable of forming the concept of THAT sort of look.16 We can either think about the
things that the appearance is presenting us with, or we can think about a particu-
lar kind of appearance, the difference being that in the former case, the reference of
our thought “reaches through” the appearances to what possesses them, while in the
latter case, we are thinking about an appearance-defined category. While observational
concepts aim to track categories that have a certain appearance but also may have an
appearance-transcendent nature, the interface that enables them to do this job also
makes available concepts of appearances themselves.
The process of learning an observational concept bears little resemblance to
traditional hypothesis-testing models of learning. For one thing, what is learned is not
a hypothesis or a judgment, but rather a two-part cognitive structure composed of a
perceptual similarity detector coupled to a symbol that it is capable of activating, but
which is also free to function autonomously in acts of higher thought. The learning

16. This point is also made by Peacocke (2001, 258): “When some property is given to a thinker
in perception in way W, then, if that thinker is sufficiently reflective, there seems to be a concep-
tual way CW of thinking of that way W, where this conceptual way CW seems to be made available
to the thinker precisely by her enjoying an experience whose nonconceptual content includes
that way W.”
240 Daniel A. Weiskopf

process itself may involve largely passive extraction of information from the environ-
ment, as when we store specific viewpoints on an object or when we make associations
between perception and action. Sometimes it may involve task-relevant but incidental
perceptual learning (Schyns 1998; Schyns, Goldstone, and Thibault 1998). Or, as in
supervised learning studies, it may involve conscious, explicit learning of categorical
distinctions, accompanied by environmental feedback.17
Nevertheless, the processes involved in building analyzers are undeniably varieties
of learning—or rather, they are if we put aside our a priori prejudices about what learn-
ing must be and focus on determining what learning as an empirical phenomenon
actually is (Laurence and Margolis 2011). The process of producing these new cognitive
structures is environmentally sensitive in such a way that the product’s detailed func-
tioning depends on its history of interactions. The representational properties of the
structures themselves reflect the kinds of inputs that shaped them. And the structures
acquired tend to be both long lasting and adaptive, in the sense that they improve the
organism’s abilities to carry out practical and cognitive tasks. These are all stereotypical
marks of learned psychological capacities (Landy and Goldstone 2005; Goldstone et al.

9.7 Observational Concepts and Conceptual Development

Once they are learned, observational concepts can resolve some longstanding puzzles
in the theory of concepts.18 One problem arises for theories that posit complex internal
structure for concepts. The constituents of a concept are its features: they pick out the
information about the concept’s referent that the concept encodes. Prototypes, exem-
plars, causal models, and other complex types of representations are built up from such
features and the relations among them. Most of the dominant models of conceptual
structure depict concepts as being composed of features (Weiskopf 2009a).
As Eleanor Rosch noted some time ago, however, this view may be developmen-
tally implausible. Describing the features elicited in some of her studies, she says:
“Some attributes, such as ‘seat’ for the object ‘chair,’ appeared to have names that
showed them not to be meaningful prior to knowledge of the object as chair” (Rosch
1978, 42). Moreover, “some attributes such as ‘you eat on it’ for the object ‘table’ were
functional attributes that seemed to require knowledge about humans, their activities,
and the real world in order to be understood” (42). The problem, in short, is that many
of the features displayed by adult concepts are ones that it is not easy to think of as

17. For discussions of the variety of category-learning tasks and the neural bases involved in each
of them, see Ashby and Maddox (2011); Segar and Miller (2010).
18. This general point is also endorsed by Roskies (2008), who argues that nonconceptual capaci-
ties are needed to explain concept acquisition. The account here aims to flesh out this picture
and to show that concepts newly learned from perception can play further bootstrapping roles in
Observational Concepts 241

being developmentally prior to the concepts that they are part of. Call this the feature
Conceptual atomists take the feature problem to be an objection to the view that
concepts are internally complex. The fact that these features are clearly linked with
concepts after they are already acquired shows that they cannot be part of those
concepts, since coming to possess a complex concept requires first possessing its parts.
Another possible interpretation, however, is that concepts are not developmentally
fixed. The initial form that a concept takes is not necessarily the same as its mature
form. Many concepts may start out life as simple observational concepts, later develop-
ing into complex, structured representations having many different constituents and
inferential liaisons.
Suppose concepts undergo a kind of developmental cycle. They might originate
as unstructured observational concepts attached to a complex perceptual analyzer.
Having an observational concept allows one to attend to objects as such—that is, to
think of them as being unified groupings that are worth keeping track of, and that
may have interesting properties and behaviors to discover. Observational concepts that
track especially useful categories, and that are frequently deployed in important cogni-
tive tasks, are then vehicles for directing our attention and intentional behavior toward
a category. The more we attend to a category, the more things we discover about its
members, including about their perceptual properties. So we may come to analyze the
appearance of a certain type of animal into its parts (legs, tail, teeth), or its observ-
able behaviors into classes (purring, grooming). As we notice these various perceivable
qualities, we can elaborate on our original observational concept and the perceptual
analyzer that underlies it. Where it might have started life as an unstructured
symbol, now it incorporates features corresponding to the various bits of information
we have collected about the typical look of the animals it responds to. The fact that
concepts undergo these types of changes in their structure and content is no surprise
if we are already committed to the pluralist idea that the conceptual system makes use
of various different types of representations in different domains, tasks, and contexts
(Weiskopf 2009b).
One aspect of conceptual development, then, might be repeated cycles of represen-
tational redescription or reencoding such as this.19 Concepts of types of objects might

19. Karmiloff-Smith (1992) also uses the term representational redescription, but my usage differs
from hers. I mean only that the very same content can be encoded in many functionally or for-
mally distinct representations at different stages of learning or development. Thus we might start
with a purely observational concept of a certain category and then later acquire a number of
exemplars from it, a prototype for it, and so on for other complex cognitive structures. These are
ways of redescribing or repackaging the same conceptual content. Karmiloff-Smith, by contrast,
means something slightly more specific, namely that content initially only grasped implicitly
will in time pass through a number of encoding stages to become both available to explicit con-
scious access and linguistic formulation.
242 Daniel A. Weiskopf

come first. An early TABLE or CAT concept might not include conceptualized information
about how those entities are put together or how they function, but this information
can come to be part of these concepts with time. Infant categorization, for instance,
often seems to be based on perceptual sensitivity to particular parts and their structural
configuration, but there is no reason to think infants conceptualize these parts until
later (Rakison and Butterworth 1998a, 1998b). From object concepts emerge concepts
of object parts, properties, and relations, so the fact that, for adults, TABLE has LEG as a
feature says nothing about developmental priority. This kind of complex developmen-
tal pattern is consistent with the emergence of more complex concepts out of simple
ones, and simple ones (ultimately) out of perception.

9.8 Conclusions

There is obviously much more to be done to flesh out the sketch of observational con-
cepts provided here. I have spoken almost entirely about perception and said nothing
about concepts that interface directly with action systems, but these are important in
developing a psychologically realistic account of basic actions. I have also not worked
out in detail the precise developmental trajectory of observational concepts, but doing
so is needed for understanding the nature of the so-called perceptual-to-conceptual
shift in infancy (Rakison 2005). Nor have I given an account of how language both
adjusts the boundaries of perceptual similarity spaces and cues the creation of new
Despite these lacunae, observational concepts seem empirically well grounded,
theoretically plausible, and indeed, arguably necessary in understanding the origins
of thought. They also suggest a larger moral. Recent debates over empiricism, and
over the extent to which thought is grounded in perception more generally, have
mostly focused on sweeping claims about the nature of all our concepts. Thoroughgo-
ing concept empiricism is unlikely to be true (Dove 2009; Machery 2007; Mahon and
Caramazza 2008; Weiskopf 2007); but empiricism might be almost true for our observa-
tional concepts. I say almost here because observational concepts are not simple copies
of percepts. As the feature problem indicates, not all of the structure in percepts makes
it into our initial observational concepts. Although this structure may be added with
time, it requires reanalysis and the construction of new representations—something
that is less characteristic of a Lockean filter, and more like a translation mechanism.
Observational concepts are not quite copies of percepts, but they are the concepts
that are shaped most powerfully by perception, and they are where the design of our
perceptual systems leaves its imprint most visibly on thought.
Claims about thought being shaped or molded by perception fall far short of
committing us to strong empiricist conclusions. On the other side of the coin,
amodal theorists also need to acknowledge that these shaping effects may, for all
Observational Concepts 243

anyone knows, be profound and far reaching. If our first thoughts involve largely
observational concepts, then this places us in a certain starting point in the space
of possible developmental trajectories for conceptual systems to follow. Where we
can go from that starting point, and what regions of conceptual space are inacces-
sible to us, depends on the richness of the processes available to us for amplifying and
refining our conceptual repertoire. Knowing the range of observational concepts we
can entertain will help to locate human thought in the space of possible conceptual
This begins to give us a sharper picture of the research directions that open up once
we adopt the notion of an observational concept. There has been an enormous amount
of developmental work focused on concept acquisition and the many mechanisms that
are implicated in it, but little consensus on how to describe the particular functional
and representational changes involved in acquiring concepts from perception. We are
now in a position to start fleshing out our sketchy understanding of how the concep-
tual system is integrated with other faculties in our cognitive architecture. Theories
of concepts have so far focused to a large extent on internal affairs: what kinds of
representations concepts are, how they are processed and deployed, and so on. These
accounts are important, and they tell us much about the organization of the concep-
tual system per se, but little about how it interfaces with the host of other cognitive
systems we possess, or how concepts as a distinctive type of representation help to
orchestrate cascades of cognitive processing. Describing the nature of these interfaces
is an open research problem, and one worth pursuing insofar as it promises to inform
us about the large-scale cognitive architecture of which our distinctively human
conceptual capacities form only a part.20


I thank the editors for their kind invitation to participate in this volume, and for their
comments on an earlier draft of this paper. Thanks to Jake Beck and Andrei Marasoiu,
who also provided excellent and detailed remarks. Special thanks to Jake for inviting me
to participate in a workshop on concept learning at Washington University in St. Louis
in May 2010 where I presented my first attempt at organizing some of these thoughts.
Finally, thanks to the organizers and participants at the workshop on Sensorimotor
Representations and Concepts (Heinrich-Heine-Universität, Düsseldorf, October 2012)
for a chance to discuss a more recent version. Some of the research and writing of this
paper took place in Spring 2012, when a Research Initiation Grant from Georgia State
University enabled me to be on leave. I gratefully acknowledge their support.

20. For some suggestions on how to distinguish higher and lower cognitive faculties, and how
conceptualized thought itself might be functionally defined, see Weiskopf (2014).
244 Daniel A. Weiskopf


Ashby, F. G., and W. T. Maddox. 2011. Human category learning 2.0. Annals of the New York
Academy of Sciences 1224:147–161.

Barsalou, L. W. 1999. Perceptual symbol systems. Behavioral and Brain Sciences 22:577–660.

Behl-Chadha, G. 1996. Basic-level and superordinate-like categorical representations in early

infancy. Cognition 60:105–141.

Behrmann, M., M. Moscovitch, and G. Winocur. 1994. Intact mental imagery and impaired
visual perception: Dissociable processes in a patient with visual agnosia. Journal of Experimental
Psychology: Human Perception and Performance 20:1068–1087.

Bukach, C. M., I. Gauthier, and M. J. Tarr. 2006. Beyond faces and modularity: The power of an
expertise framework. Trends in Cognitive Sciences 10:159–166.

Carey, S. 2009. The Origin of Concepts. New York: Oxford University Press.

Clark, A. 2004. Feature-placing and proto-objects. Philosophical Psychology 17:443–469.

Dahlgrun, M. 2011. The notion of a recognitional concept and other confusions. Philosophical
Studies 150:139–160.

De Renzi, E., and F. Lucchelli. 1994. Are semantic systems separately represented in the brain?
The case of living category impairment. Cortex 30:3–25.

Dixon, M. J., G. Desmarais, C. Gojmerac, T. A. Schweizer, and D. Bub. 2002. The role of premor-
bid expertise on object identification in category-specific visual agnosia. Cognitive Neuropsychology

Dove, G. 2009. Beyond perceptual symbols: A call for representational pluralism. Cognition

Dretske, F. 1986. Explaining Behavior. Cambridge, MA: MIT Press.

Eimas, P. D., and P. C. Quinn. 1994. Studies on the formation of perceptually based basic-level
categories in young infants. Cognitive Development 65:903–917.

Farah, M. 2004. Visual Agnosia. 2nd ed. Cambridge, MA: MIT Press.

Feldman, J. 2003. What is a visual object? Trends in Cognitive Sciences 7:252–256.

Fodor, J. 1998. There are no recognitional concepts, not even RED. Philosophical Issues 9:1–14.

Fodor, J. 2008. LOT 2: The Language of Thought Revisited. Oxford: Oxford University Press.

Gauker, C. 2011. Words and Images. Oxford: Oxford University Press.

Gauthier, I., M. J. Tarr, A. W. Anderson, P. Skudlarski, and J. C. Gore. 1999. Activation of the
middle fusiform “face area” increases with expertise recognizing novel objects. Nature Neurosci-
ence 2:568–573.
Observational Concepts 245

Goldstone, R. L., A. Gerganov, D. Landy, and M. E. Roberts. 2008. Learning to see and conceive.
In The New Cognitive Sciences, ed. L. Tommasi, M. Peterson, and L. Nadel, 163–188. Cambridge,
MA: MIT Press.

Henderson, J. M., and A. Hollingworth. 1999. High-level scene perception. Annual Review of
Psychology 50:243–271.

Humphreys, G. W., and M. J. Riddoch. 1987. To See but Not to See: A Case Study of Visual Agnosia.
Hillsdale, NJ: Erlbaum.

Humphreys, G. W., and M. J. Riddoch. 2006. Features, objects, action: The cognitive neuropsy-
chology of visual object processing, 1984–2004. Cognitive Neuropsychology 23:156–183.

Humphreys, G. W., and M. J. Riddoch. 2007. How to define an object. Mind & Language

Hurley, S. 2008. The shared circuits model (SCM): How control, mirroring, and simulation can
enable imitation, deliberation, and mindreading. Behavioral and Brain Sciences 31:1–22.

Jacob, P., and M. Jeannerod. 2003. Ways of Seeing: The Scope and Limits of Visual Cognition. Oxford:
Oxford University Press.

Karmiloff-Smith, A. 1992. Beyond Modularity: A Developmental Perspective on Cognitive Science.

Cambridge, MA: MIT Press.

Lamme, V. A. F. 2006. Towards a true neural stance on consciousness. Trends in Cognitive Sciences

Lamme, V. A. F., and P. R. Roelfsema. 2000. The distinct modes of vision offered by feedforward
and recurrent processing. Trends in Neurosciences 23:571–579.

Landy, D., and R. L. Goldstone. 2005. How we learn about things we don’t already understand.
Journal of Experimental & Theoretical Artificial Intelligence 17:343–369.

Laurence, S., and E. Margolis. 2011. Learning matters: The role of learning in concept acquisition.
Mind & Language 26:507–539.

Lyons, W. 2008. Perception and Basic Beliefs. Oxford: Oxford University Press.

Machery, E. 2007. Concept empiricism: A methodological critique. Cognition 104:19–46.

Mahon, B., and A. Caramazza. 2008. A critical look at the embodied cognition hypothesis and a
new proposal for grounding conceptual content. Journal of Physiology, Paris 102:59–70.

Mandler, J. 2004. The Foundations of Mind. Oxford: Oxford University Press.

Mandler, J. M., and L. McDonough. 1993. Concept formation in infancy. Cognitive Development

Mandler, J. M., and L. McDonough. 1998. On developing a knowledge base in infancy. Develop-
mental Psychology 34:1274–1288.
246 Daniel A. Weiskopf

Matthen, M. 2005. Seeing, Doing, and Knowing. Oxford: Oxford University Press.

Milner, A. D., and M. A. Goodale. 2006. The Visual Brain in Action. 2nd ed. Oxford: Oxford
University Press.

Peacocke, C. 1992. A Study of Concepts. Cambridge, MA: MIT Press.

Peacocke, C. 2001. Does perception have a nonconceptual content? Journal of Philosophy


Pessig, J. J., and M. J. Tarr. 2007. Visual object recognition: Do we know more now than we did 20
years ago? Annual Review of Psychology 58:75–96.

Prinz, J. 2002. Furnishing the Mind. Cambridge, MA: MIT Press.

Prinz, J. 2005. Beyond appearances: The content of sensation and perception. In Perceptual Experi-
ence, ed. T. Gendler and J. Hawthorne, 434–460. Oxford: Oxford University Press.

Prinz, W. 1997. Perception and action planning. European Journal of Cognitive Psychology 9:

Pylyshyn, Z. W. 2001. Visual indexes, preconceptual objects, and situated vision. Cognition

Quinn, P. C., P. D. Eimas, and S. Rosenkrantz. 1993. Evidence for representations of

perceptually similar natural categories by 3-month-old and 4-month-old infants. Perception 22:

Rakison, D. H. 2005. The perceptual to conceptual shift in infancy and early childhood: A surface
or deep distinction? In Building Object Categories in Developmental Time, ed. L. Gershkoff-Stowe
and D. H. Rakison, 131–158. Mahwah, NJ: Erlbaum.

Rakison, D. H., and G. E. Butterworth. 1998a. Infants’ attention to object structure in early
categorization. Developmental Psychology 34:1310–1325.

Rakison, D. H., and G. E. Butterworth. 1998b. Infants’ use of object parts in early categorization.
Developmental Psychology 34:49–62.

Recanati, F. 2002. The Fodorian fallacy. Analysis 62:285–289.

Riesenhuber, M., and T. Poggio. 2000. Models of object recognition. Nature Neuroscience 3
(supplement): 1199–1204.

Rives, B. 2010. Concepts and perceptual belief: How (not) to defend recognitional concepts. Acta
Analytica 25:369–391.

Rosch, E. 1978. Principles of categorization. In Cognition and Categorization, ed. E. Rosch and
B. Lloyd, 27–48. Hillsdale, NJ: Erlbaum.

Rosch, E., C. B. Mervis, W. D. Gery, D. M. Johnson, and P. Boyes-Braem. 1976. Basic objects in
natural categories. Cognitive Psychology 8:382–439.
Observational Concepts 247

Roskies, A. 2008. A new argument for nonconceptual content. Philosophy and Phenomenological
Research 76:633–659.

Scholl, B. 2001. Objects and attention: The state of the art. Cognition 80:1–46.

Schütz-Bosbach, S., and W. Prinz. 2007. Perceptual resonance: Action-induced modulation of

perception. Trends in Cognitive Sciences 11:349–355.

Schyns, P. 1998. Diagnostic recognition: Task constraints, object information, and their interac-
tions. Cognition 67:147–179.

Schyns, P., R. Goldstone, and J. Thibault. 1998. The development of features in object concepts.
Behavioral and Brain Sciences 21:1–54.

Segar, C. A., and E. K. Miller. 2010. Category learning in the brain. Annual Review of Neuroscience

Siegel, S. 2010. The Contents of Visual Experience. Oxford: Oxford University Press.

Tanaka, J. W. 2001. The entry point of face recognition: Evidence for face expertise. Journal of
Experimental Psychology. General 130:534–543.

Tanaka, J. W., and I. Gauthier. 1997. Expertise in object and face recognition. Psychology of
Learning and Motivation 36:83–125.

Tarr, M. J. 2003. Visual object recognition: Can a single mechanism suffice? In Perception of Faces,
Objects, and Scenes: Analytic and Holistic Processes, ed. M. A. Peterson and G. Rhodes, 177–211.
Oxford: Oxford University Press.

Tye, M. 2005. Nonconceptual content, richness, and fineness of grain. In Perceptual Experience, ed.
T. Gendler and J. Hawthorne, 504–530. Oxford: Oxford University Press.

Weiskopf, D. A. 2007. Concept empiricism and the vehicles of thought. Journal of Consciousness
Studies 14:156–183.

Weiskopf, D. A. 2009a. Atomism, pluralism, and conceptual content. Philosophy and Phenomeno-
logical Research 79:130–162.

Weiskopf, D. A. 2009b. The plurality of concepts. Synthese 169:145–173.

Weiskopf, D. A. 2014. The architecture of higher thought. In New Waves in Philosophy of Mind, ed.
M. Sprevak and J. Kallestrup, 242–261. New York: Palgrave Macmillan.
V Concepts and Language
10 What’s in a Concept? Analog versus Parametric Concepts
in LCCM Theory

Vyvyan Evans

10.1 Introduction

Any account of language and the human mind has to grapple, ultimately, with
the nature of concepts. In the words of cognitive scientist Jesse Prinz, concepts are
“the basic timber of our mental lives” (Prinz 2002, 1). For without concepts, there
could be no thought, and language would have nothing to express. What is less
clear, however, is exactly how concepts are constituted, and the relationship between
language and concepts. These are the two issues I address in this chapter. I do so by
posing and attempting to answer the following question: Do linguistic units (e.g.,
words) have semantic content independently of the human conceptual system (which,
in rough terms, can be thought of as our repository of concepts)? The answer to this
question is, I will argue, a clear yes.
The thrust of the argument I present in this chapter is that there is a qualitative
difference between concept types. Concepts in the conceptual system, on the one
hand, and in the linguistic system, on the other, are of two qualitatively different
sorts, which reflect the function of the two systems. The conceptual system is, in evo-
lutionary terms, far older than the linguistic system. And at least in outline, many
other species have conceptual systems that are continuous with the human conceptual
system. In contrast, language evolved, I argue, to provide an executive control func-
tion, harnessing concepts in the conceptual system for purposes of (linguistically medi-
ated) communication. The consequence is that the concepts that inhere in each of
these systems evolved to fulfill distinct, albeit complementary purposes. Moreover, the
findings giving rise to a grounded (or embodied) cognition perspective has, in recent
years, led to a reframing of how we should think about concepts, in both the concep-
tual and the linguistic systems.
Accordingly, in this chapter I present arguments for thinking that the distinction
between the conceptual and linguistic systems relates to concepts that are analog in
nature, on the one hand, and those that are parametric in nature, on the other. In so
doing, I argue against received accounts of embodied cognition that fail to recognize
252 Vyvyan Evans

such a distinction. I also argue against disembodied accounts of concepts. My overall

conclusion is that parametric concepts facilitate access to analog concepts in the process
of meaning construction. Although both types of concept are derived from embodied,
or as I shall prefer, grounded experience, they are qualitatively distinct. Parametric
concepts are schematic, while analog concepts are richer, more closely constituting
analogs of the experience types they are grounded in. Once I have developed an
account of these distinct concepts, I advance a theory of lexical representation and
semantic composition, referred to as the theory of lexical concepts and cognitive
models, or LCCM theory for short (Evans 2006, 2009, 2010, 2013). I use LCCM the-
ory to show the distinct functions of parametric and analog concepts in meaning
The chapter is organized as follows. In the next section, I advance a grounded cog-
nition approach to lexical and conceptual representation. In particular, I argue that
representations in the conceptual system are multimodal, being constituted of a range
of information types, including sensorimotor information. I also disambiguate knowl-
edge representation (concepts) from meaning construction (semantic composition).
The two phenomena are often conflated in the cognitive science literature on concepts.
But, I argue, they are, in fact, distinct. In section 10.3 I review evidence for think-
ing that language is subserved by analog concepts (within the conceptual system). I
do so by reviewing empirical findings for what I term grounding effects. Section 10.4
then develops the central qualitative distinction I argue for, between nonlinguistic
and linguistic concepts. I do so by first observing that extant embodied/grounded
theories of concepts assume that linguistic meaning is equivalent to conceptual
representation. I present arguments that lead to a distinction in terms of conceptual
versus lexical representations. I then operationalize this distinction in terms of
parametric concepts (linguistic system) versus analog concepts (conceptual system).
Section 10.6 introduces LCCM theory, which operationalizes these distinct types
of representation in terms of the theoretical constructs of lexical concept (linguistic
system) and cognitive model (conceptual system). The LCCM theory framework
provides a basis for understanding the respective contribution of each distinct
concept type in facilitating linguistically mediated meaning construction, which
is complementary, albeit orthogonal to, an account of knowledge representation

10.2 Toward a Grounded Cognition Approach to Concepts

Broadly speaking, there are, within cognitive science, two prevalent views of concepts.
The first view is, very roughly, that concepts are abstract, disembodied symbols—
Barsalou (1999, 2008) describes this perspective as the amodal view of cognition—
What’s in a Concept? 253

a view that assumes that the representational format of a concept is qualitatively dif-
ferent from the sensory experiences concepts relate to. Although the details of spe-
cific disembodied theories vary considerably, this general perspective assumes that
concepts are ultimately abstracted from the brain states that give rise to them, with
information encoded in a different format. This view of concepts makes a principled
distinction between conception (or cognition) and perception (and interoceptive
experience more generally)—see Cisek (1999) for discussion. Representatives of this
general approach include Dennett (1969), Fodor (1975, 1983, 1998, 2008), Hauge-
land (1985), Jackendoff (1983, 1987), Newell and Simon (1972), Pinker (1984), and
Pylyshyn (1984).
More recently, a different perspective has emerged, which blurs the distinction
between perception/interoception and conception/cognition. On this view, concepts
are directly grounded in the perceptual and interoceptive brain states that give rise
to them. Again, while details relating to specific theories differ, this embodied, modal,
or as I prefer, grounded cognition perspective sees cognition as broadly continuous
with perception/interoception, rather than reflecting a wholly distinct type of repre-
sentation (see Barsalou 2008 and Shapiro 2010 for reviews). Notable exemplars of this
view include Barsalou (e.g., 1999), Chemero (2009), Clark (e.g., 1997), Damasio (1994),
Evans (2009), Gallese and Lakoff (2005); Glenberg (e.g., 1997), Lakoff and Johnson
(e.g., 1999), Viglioccio et al. (2009), and Zwaan (e.g., 2004).
The grounded view assumes that concepts arise directly from representative brain
states. Take the example of the experience of cats. When we perceive and interact with
cats, this leads to extraction of perceptual and functional attributes of cats, which are
stored in memory in analog fashion: our concept CAT, on this view, closely resembles
our perception and experience of a cat. When we imagine a cat, this is made possible
by reactivating, or to use the technical term, simulating the perceptual and interocep-
tive experience of interacting with a cat—these include sensorimotor experiences when
we stroke and otherwise interact with a cat, as well as affective states, such as the
pleasure we experience when a cat responds by purring, and so forth. But while the
simulated cat closely resembles our conscious perceptual and interoceptive experience,
it is attenuated.
In other words, the concept CAT is not the same as the veridical experience of per-
ceiving a cat. When we close our eyes and imagine a cat, we are at liberty to simulate an
individual cat, or a type of cat, or a cat composed of aspects of our past experiences of
and with cats. But the simulation is attenuated with respect to the veridical perceptual
experience of a cat. Importantly, the claim made by the embodied perspective is that
the simulation is directly grounded in the same brain states—in fact, a reactivation of
aspects of the brain states—that are active when we veridically perceive and interact
with the cat. The simulation is then available for language and thought processes. As
254 Vyvyan Evans

the reactivation of some aspects of the perceptual and interoceptive experiences of a cat
is, in part, constitutive of the concept CAT, the concept is an analog of the perceptual
In contrast, the disembodied view of concepts and mind assumes that perceptual
experiences are redescribed into a symbol, which stands for, or tokens, the perceptual
experience. In some disembodied theories, the symbols are represented using natural
language, and the symbols are thought to comprise lists of features or attributes. In
others (e.g., Fodor 1975, 2008), the concepts are represented in a format that is in some
sense language-like: the idea is that the mind features its own operating system, univer-
sal to all humans—mentalese. Various approaches to mentalese have been developed
in some detail (see, e.g., Fodor 2008; Jackendoff 1983).
The key difference between the two perspectives is that the disembodied view of
concepts assumes that concepts are mental representations fundamentally unlike what
they represent. Thus, critically, perceptual and interoceptive brain states are not consti-
tutive of concepts. For embodied cognition proponents, simulations, in contrast, are
analog presentations, in the sense of re-presentations of perceptual and interoceptive
experiences—they are directly grounded in the body-based and subjective perceptual
states that give rise to them. As such the grounded cognition view assumes that per-
ceptual and interoceptive brain states are constitutive of concepts. Figures 10.1 and
10.2 capture the distinctions between the disembodied and grounded approaches to
A particular challenge that has been leveled against the disembodied view of
concepts relates to what has been dubbed the “symbol grounding problem” (Harnard
1990). What is at stake is the nature of content available for semantic analysis, given
that concepts presumably facilitate thought and language—about which I shall have
more to say in section 10.3. In other words, if symbols are abstract, which is to say,
unlike the perceptual and interoceptive mental states they represent, how do they
relate to the states they are supposed to be representative of? In short, the challenge for

CAT = C1
Fur = f1
Paws = p1

Perceptual/interoceptive experience Feature list symbol stored in memory

Figure 10.1
Disembodied conceptual system.
What’s in a Concept? 255




Perceptual/interoceptive experience Analogue (albeit attenuated)

representation stored in memory

Figure 10.2
Grounded conceptual system.

the disembodied view is to show how concepts facilitate semantic analysis when they
are not directly grounded in the content that they represent.
One early solution to this, in one disembodied theory of concepts, was to assume
that concepts are innate. This was the proposal made by Fodor (1975), detailing men-
talese (a.k.a. the language of thought). Adopting this view, Jackendoff posited a range
of abstract conceptual primitives that could be combined using rules of mental syntax,
facilitating a full-fledged disembodied conceptual system. In later work, Fodor (e.g.,
1998) recanted, arguing that concepts were not themselves innate. Rather, concepts
are grounded by virtue of a process whereby abstract symbols became locked to the
perceptual states in the world that they represent, or token, in his parlance. He declined
to speculate on how this locking process comes about, however, arguing that although
the mechanisms that facilitate it may be innate, understanding how symbols become
locked to perceptual states is not the proper concern of cognitive psychology.
The grounding problem is resolved in the grounded cognition view of concepts
by positing that concepts are directly grounded in brain states: hence concepts are
very much like the brain states they are representations of. Most versions of grounded
cognition (e.g., Barsalou 1999; Clark 1997) assume that although brain states are con-
stitutive of concepts, concepts are nevertheless representations, and thus distinct from
brain states (e.g., sensorimotor activations). Barsalou (1999) refers to such concepts as
“perceptual symbols”: on this view, perceptual symbols are stored traces of prior brain
states that can be reactivated, or simulated, for purposes of language and thought.
Hence, the difference from the disembodied view is that perceptual symbols are directly
grounded in brain states.
A more radical approach to embodied cognition removes the requirement for
symbols altogether. For instance, Chemero (2009) argues that concepts are entirely
constituted of brain states (rather than representations of prior brain states). Build-
ing on dynamic systems theory, and James Gibson’s ecological approach to
256 Vyvyan Evans

perception, Chemero thereby entirely removes the distinction between perception

and conception.
Two main lines of empirical evidence support the grounded cognition view of

1. Brain-based demonstrations that the sensorimotor and other modal systems are acti-
vated during conceptual processing: a raft of studies provides clear evidence that, for
instance, motor processes are automatically engaged when subjects perform perceptual
and conceptual tasks. A range of different methodologies have demonstrated such auto-
matic activation in both nonhuman primates and humans. For instance, the human
motor system is automatically activated when subjects observe manipulable objects,
and when they observe the actions of another individual. Methodologies deployed
to demonstrate such data include studies involving functional neuroimaging, neuro-
physiological recordings, EEG, transcranial magnetic stimulation (TMS), and kinematic
analyses. For representative reviews of the various extant findings, see Barsalou (2008),
Boulenger et al. (2008), Gallese and Lakoff (2005), Pulvermüller et al. (2005), and
Rizzolatti and Craighero (2004).
2. Behavioral demonstrations that activation of the sensorimotor system spread to
conceptual levels of processing. Many of the relevant studies have involved sentence
comprehension and lexical decision tasks. I will have more to say about the relation-
ship between language and concepts below; however, one representative and impor-
tant study is Pulvermüller et al. (2005). For instance, subjects were required to perform
a lexical decision task employing action verbs relating to either arm or leg actions. A
pulse of subthreshold TMS was then applied to either the leg or arm region of motor
cortex immediately after exposure to the lexical cues. Pulvermüller and colleagues
found that TMS exposure to the arm region induced a faster decision for arm action
words, as opposed to leg action words. And when TMS exposure was to the leg region
of the motor cortex, the reverse pattern in lexical decisions was found. This would
appear to suggest that embodied (perceptual and interoceptive) states play a direct role
in conceptual processing as it relates to language comprehension.

In the light of demonstrations such as these, the difficulty for a disembodied view
of concepts—at least as classically formulated—is this: concepts are supposed to be
abstract symbols, which are not constituted by sensorimotor (and interoceptive) brain
states. In short, semantic analysis of concepts is independent of sensorimotor and other
brain states, and hence, should not result in automatic processing of these. Although
disembodied accounts would not deny that concepts for sensorimotor experience, for
instance, will require activation of the relevant region of sensorimotor cortex, the point
is that such activations are ancillary to the semantic analysis that the disembodied
symbols furnish, for purposes of facilitating language and thought. That is, sensorimo-
tor activation plays no role in semantic analysis during conceptual processing. Hence,
What’s in a Concept? 257

this finding—that brain states such as sensorimotor activations appear to play a role in
conceptual processing—would appear to falsify disembodied accounts of concepts, as
classically formulated.1
It is also worth noting that just as the disembodied view is falsified by empirical
findings briefly reviewed above, so too is the radical view of embodied cognition. The
relevant finding concerns patients with apraxia—patients with impairments for using
objects such as hammers are nevertheless able to name and describe in detail the nature
of the same objects they cannot use (Johnson-Frey 2004; Mahon and Caramazza 2008.
This illustrates that subjects appear to have some type of representation, at least for
purposes of linguistic encoding, without being able to activate a supporting sensorimo-
tor perceptual state. This finding would appear to undermine the radical claim that
representations are not required for a functioning conceptual system.
The empirical findings about the role of brain states, especially sensorimotor activa-
tion in conceptual processing, appear to demonstrate that concepts, especially senso-
rimotor concepts, involve activation of brain states. However, this is unlikely to be the
whole story: Mahon and Caramazza (2008) argue that part of the disembodied account,
that concepts involve abstract (and hence non-analog) symbols, appears to be sup-
ported by findings such as patients with brain motor deficits who nevertheless retain
motor concepts, as in the case of apraxia. Accordingly, they have proposed an account
of concepts that can be seen as situated midway between an embodied and a disem-
bodied account of concepts, integrating what they perceive to be the strengths of each.
Mahon and Caramazza accept the empirical findings that have been used to argue for
a grounded view of concepts. However, they argue that such data do not entail that a
disembodied view is incorrect.
First, they argue that any embodied view is limited up front because it assumes that
all concepts are constituted by sensorimotor experience. And that being so, an embod-
ied approach inevitably cannot deal with abstract concepts such as TIME, JUSTICE, ENTROPY,
or PATIENCE. For some of these concepts, they argue, there doesn’t appear to be senso-
rimotor information that could correspond in any reliable or direct way to the abstract
concept. For others, such as BEAUTIFUL, there appears not to be consistent sensorimotor
experience associated with it. For instance, mountains, a lover’s face, or even an idea
can be beautiful: no specific type of sensorimotor information is apparently instru-
mental in the realization of the concept BEAUTIFUL. Second, some information is clearly
abstract even for sensorimotor concepts, as evidenced in the case of apraxia, already

1. Some commentators have observed, however, that a suitably modified version of the disem-
bodied account may be consistent with data that have led to the embodied/grounded cognition
accounts—for discussion see Dove (2009), Mahon and Caramazza (2008), and Machery (2007).
For criticism of the embodied approach on philosophical grounds, see Weiskopf (2010). I discuss
proposals made by Mahon and Caramazza below.
258 Vyvyan Evans

discussed, where part of the concept is retained in the absence of the ability to simulate
the motor experience associated with the concept.
In light of these observations, Mahon and Caramazza posit a “grounding by interac-
tion” view of concepts. On this account, they retain the idea that concepts consist of
abstract symbols—symbols that are non-analog in nature. Hence, they argue, concepts
are not constituted by the brain’s modal states, such as sensory and motor modali-
ties. As they accept the empirical findings of the embodiment tradition, however, they
assume that embodied states are not ancillary to conceptual processing. Rather, what
they claim is that modality-specific activation of brain states, such as sensorimotor
information, may constitute instantiations of the concept. To illustrate, reconsider the
concept HAMMER. For Mahon and Caramazza, the concept consists of an abstract, disem-
bodied symbol. And this disembodied symbol is the relevant unit for semantic analysis
(upon which language and thought are contingent). That said, the referent of the
concept is a concrete entity in veridical experience. And this is instantiated as senso-
rimotor experience. Hence, on this account, the abstract symbol HAMMER is grounded by
virtue of an interaction between the abstract symbol and sensorimotor activation: they
propose that the sensorimotor activation results from a cascading spreading activation
from the conceptual system—where the concept is housed—to the relevant region of
the motor cortex.
There are a number of problems with the grounding by interaction view. First, the
assumptions it makes about the embodied cognition view are erroneous. To argue that
embodied cognition researchers assume that all concepts are constituted exclusively
of sensorimotor experience is patently incorrect. For instance, Barsalou’s account of
perceptual symbols, arguably the best-developed theoretical account of concepts from
an embodied cognition perspective, explicitly includes brain states other than senso-
rimotor experience. Barsalou uses the term perception in a rather more inclusive way
than is normally the case: he assumes that perceptual symbols include sensorimotor
experiences as well as other sorts of more subjective modalities, such as interoceptive
experience, affect, and cognitive states. In later work, Barsalou (2008) explicitly prefers
the term grounded—rather than embodied—cognition in order to make clear that, in
his view, concepts encompass all of the brain’s modal systems, not just sensorimotor
areas. Grounded cognition is also the nomenclature preferred in the present chapter.
Second, Mahon and Caramazza are incorrect in claiming that the embodied cogni-
tion view cannot account for abstract concepts up front. An entire research tradition
associated with Lakoff and Johnson (e.g., 1980, 1999) has argued that abstract concepts
are indeed largely constituted in terms of sensorimotor experience. Take for instance,
the abstract concept TIME. Lakoff and Johnson argue that the domain of time is sys-
tematically structured by mappings, which inhere in long-term memory, that provide
a long-term stable link from the domain of motion through space. The series of map-
pings that facilitate the projection of conceptual structure, asymmetrically from one
What’s in a Concept? 259

domain onto another, are termed conceptual metaphors. These conceptual metaphors
facilitate the structuring of concepts such as DURATION in terms of physical extent (i.e.,
length), as implied by linguistic examples such as (1):

(1) The relationship lasted a long time.

In (1), long refers to temporal rather than spatial extent (see Evans 2013 for discus-
sion). And behavioral findings support the view that spatial extent is automatically
activated by the concept of duration but not vice versa (Casasanto and Boroditsky
2008). Conceptual metaphor accounts exist for a host of abstract domains ranging
from mathematics (Lakoff and Nuñez 2001) to morality (Lakoff and Johnson 1999).
And behavioral studies now provide evidence that abstract concepts involve process-
ing of sensorimotor experience—about which I shall have more to say in the next
That said, what has perhaps been underestimated by embodied cognition research-
ers, including Lakoff and Johnson, is the degree to which abstract concepts are consti-
tuted by information from modalities other than the sensory and motor modalities.
With respect to time, for instance, at the neurological level, it appears that the range
of phenomenologically real temporal experiences, such as duration, succession, and
present/past/future, are subserved by a range of non-sensorimotor systems (for reviews,
see Evans 2013; Kranjec and Chatterjee 2010). Nevertheless, representations for time
also appear to be constituted, in part, in terms of sensorimotor information. Accounts
differ as to the reason for this (Bonato et al. 2010 versus Bueti and Walsh 2009). But
the parietal cortex appears to be implicated in linking representations for space
and time.
The third problem with the grounding by interaction view is that accounting for
automatic activation of sensorimotor information by allowing for interaction still
doesn’t provide a grounded theory of concepts. After all, the sensorimotor information
still remains ancillary, as per disembodied theories: semantic analysis takes place with-
out reference to the sensorimotor information. In essence, Mahon and Caramazza’s
proposal remains essentially a disembodied account, but with bells and whistles: it is
arguably consistent with the empirical evidence, but the grounding problem is still
not resolved because modality-specific information remains excluded from concep-
tual content. An alternative, which I explore below, is that concepts are of different
types, some directly grounded in perceptual and interoceptive experience—what I refer
to as analog concepts—and some that represent abstractions derived from grounded
experience—parametric concepts. As we shall see, parametric concepts are abstract; but
they are grounded in modal brain states by virtue of being schematizations, abstracted
from said brain states.
The final problem is that Mahon and Caramazza conflate, and confuse, knowledge
representation—the issue of what constitutes a concept—and meaning construction.
260 Vyvyan Evans

Their discussion of the concept BEAUTIFUL and the divergent sensorimotor properties
instantiated by this concept requires an account of meaning construction, rather than
knowledge representation. Meaning construction is, in part, a linguistically mediated
phenomenon, and it requires an appropriate account of compositional semantics. An
account of the concept BEAUTIFUL in different contexts of use turns on issues relating
to semantic composition and language use, rather than knowledge representation. It
is therefore disingenuous to criticize embodied cognition accounts of concepts when
pointing to phenomena that relate to something other than knowledge representation.
Once I have developed my account of analog versus parametric concepts, I present,
later in the chapter, an account of semantic composition within the framework of
LCCM theory, which addresses this issue.
In the final analysis, concepts appear to be constituted, in part, by multimodal brain
states, not exclusively sensorimotor experience types. A theory of conceptual represen-
tation is, in principle, distinct from that of linguistically mediated semantic composi-
tion. And as such, we must now turn to a consideration of the relationship between
concepts and language.

10.3 Grounding Effects in Language

If concepts subserve language, then it stands to reason that language relies on concepts,
at least in part, to facilitate the construction of meaning. The purpose of this section
is to review the evidence in support of such a view. Given the grounded cognition
perspective developed above, if language is subserved by concepts grounded in multi-
modal brain states, we should find evidence of grounding effects in language. A ground-
ing effect, as I define it, constitutes an observable and intersubjectively robust intrusion
of embodied (not exclusively sensorimotor) experience in conceptual and especially
linguistic representation and processing.
Recent findings in both psychology and cognitive neuroscience now clearly reveal
a series of grounding effects when we use language (see table 10.1). First, multimodal
brain states are activated when we deploy language. Moreover, these activations are

Table 10.1
Three types of grounding effects in linguistic cognition

Automatic activation of Brain regions that are specialized for specific types of processing
brain regions are activated when the corresponding language is processed.
Immersed bodily behavior Specialized bodily behaviors are activated by the processing of
the corresponding language.
Structure and organization Language appears to be structured in terms of embodied brain
of language states, especially representations grounded in sensorimotor
What’s in a Concept? 261

fast—multimodal information is activated instantaneously, automatically, and can-

not be avoided—and somatotopic—they relate to specific functional brain regions;
for instance, action words referring to hand actions activate the hand area of motor
cortex and not the leg area (Pulvermüller et al. 2005). Second, psychologists have dis-
covered that human subjects behave as if immersed in an embodied state when using
or understanding language relating to perceptual experience. Third, grounding effects
show up directly in the nature and structure of language. Together, this amounts to
persuasive evidence in favor of the grounded view I am advancing: the human mind is
continuous with the human body and bodily experience, rather than being a separate
So what then are examples of grounding effects? Let’s focus on the somatotopic
aspect of brain activation: specific brain regions are activated when we use the corre-
sponding words, or types of words. It is now well established that distinct parts of the
cortex process and store sensory information: for instance, visual, auditory and tactile
experience. Other parts of the cortex process motor information: for instance, informa-
tion relating to hand or body movements. And finally, subcortical structures, such as
the amygdala, process and store emotional experience. Recent findings have shown
that all of these brain regions are automatically and immediately activated when
corresponding body-based language is being processed.
For example, brain regions that are active during the processing of actions, such
as using tools like hammers, screwdrivers, and saws, are automatically and immedi-
ately activated when we hear or read sentences relating to using tools of these kinds
(Isenberg et al. 1999; Martin and Chao 2001; Pulvermüller 1999; see also Buccino
et al. 2005; for a review, see Taylor and Zwaan 2009). Put another way, when you
or I understand an expression such as “He hammered the nail,” there is automatic
and immediate activation of that part of the brain that is engaged to produce the
hammering action. In addition, regions of the brain that process visual information are
activated when we comprehend words and sentences relating to visual information,
such as object shape and orientation (Stanfield and Zwaan 2001; Zwaan and Yaxley
2003). For instance, visual areas that process animal recognition shapes are activated
when we hear or see certain animal words (Büchel, Price, and Friston 1998; Martin and
Chao 2001). And finally, language involving emotional affect also results in automatic
activation of the relevant brain regions. For instance, threat words such as destroy and
mutilate automatically activate parts of the amygdala (Isenberg et al. 1999). This is an
evolutionarily older part of the subcortical brain that neurobiologists have established
as being involved in emotional processing (LeDoux 1995).
The second type of grounding effect is behavior. Human subjects, when using
or understanding language, behave in myriad subtle ways, as if they are engaged
in the sensorimotor activity that corresponds to the sensorimotor language; it is as
if language primes language users for particular veridical actions. For instance, when
262 Vyvyan Evans

reading about throwing a dart in a game of darts, human subjects automatically acti-
vate muscle systems that ready the hand grip common to dart throwing; when we
use or hear language, our eye and hand movements are consistent with the sensorimo-
tor activity being described (Glenberg and Kaschak 2002; Klatzky et al. 1989; Spivey
et al. 2000). It is as if language facilitates the vicarious experience of the events being
described in language.
The psycholinguist Rolf Zwaan has described this in terms of language users being
immersed experiencers. He argues that “language is a set of cues to the comprehender
to construct an experiential (perception plus action) simulation of the described situa-
tion” (Zwaan 2004, 36). And this could only be so if language provides direct access to
representations of body-based states: concepts are embodied.
Behavioral evidence for immersion in embodied states, when using language, comes
from the psychology lab. In one experiment, subjects were asked to judge whether
action sentences such as “He closed the drawer” were meaningful or not (Glenberg and
Kaschak 2002). Subjects did this by pressing one of two buttons, which were located
sequentially in front of the subject. The button signaling that a sentence was mean-
ingful was closer to the subjects and thus involved moving their hand toward their
body, the same direction of motor control required to open a drawer. It was found that
responses to whether the sentences were correct or not were faster when the direction
of motion corresponded to that described in the sentence. This finding supports the
view that bodily motor states are automatically activated when reading a correspond-
ing sentence. An action required by the experiment that is at odds with the motor
simulation activated by the sentence provides interference. And this, in turn, slows
down the subject’s response to the semantic judgment, the ostensible purpose of the
The third type of grounding effect derives from the structure and organization of
language: language for abstract states appears to draw on language from sensorimotor
experience in an asymmetric way. Linguistic evidence of this sort is compatible with
the grounded cognition view of concepts but not the disembodied perspective. Perhaps
the most clear evidence in language has been highlighted in the work of Lakoff and
Johnson (1980, 1999). As noted in the previous section, conceptual metaphors appear
to work by recruiting structure from sensorimotor experience in order to structure
representations relating to interoceptive experience types. For instance, various aspects
of our representations for time appear to be systematically structured in terms of
representations recruited from the domain of (motion through) space. Consider some
linguistic examples, which have been claimed to evidence this:

(2a) Christmas is fast approaching.

(2b) We are moving up on Christmas fast.

What’s in a Concept? 263

These examples suggest the following. The relative imminence of a future event,
Christmas, is structured in terms of the motion of an event—an event conceptualized
as if it were an object capable of motion—toward the ego, or the ego’s motion toward
Christmas, conceived as a location in space. Lakoff and Johnson posit that we structure
our representations of time in terms of relative motion of objects or our relative motion
with respect to stationary objects (see also Moore 2006). Moreover, the evidence for
conceptual metaphors—from language, from psycholinguistic tasks (Boroditsky 2000),
and from psychophysical tasks (Casasanto and Boroditsky 2008)—appears to show
that the structuring is asymmetric. That is, representations for time are systematically
structured in terms of representations for space and motion through space, but space
appears not to be productively structured in terms of representations for time.
Interestingly, and in keeping with the proposals made by Lakoff and Johnson, a
range of abstract concepts also appear to exhibit grounding effects. Lakoff and Johnson
have argued that we conceptualize communication as physical transfer. Evidence for
this comes from linguistic examples, as when we say things like, “I couldn’t get my
ideas across,” “put it into words,” and so on. Indeed, Glenberg and Kaschak (2002)
found that the same pattern applied to abstract concepts.
Consider a sentence such as “I gave him some words of wisdom.” Metaphorically,
this involves transferring the “words of wisdom,” some advice, from the speaker to
the listener, a pattern of motion away from the body. The processing time to judge
whether the sentence was semantically acceptable was quicker when the button that
was required to be pressed involved an action away from rather than toward the
subjects’ bodies. In other words, physical action that accorded with the metaphorical
action facilitated faster understanding of the linguistic expression. What this reveals is
a grounding effect for abstract, as well as literal, language, a finding in keeping with the
broad prediction of conceptual metaphor theory.
Further evidence for abstract concepts being structured, at least in part, by senso-
rimotor experience, comes from the work of Casasanto and Dijkstra (2010). In one
experiment, Casasanto and Dijkstra investigated abstract concepts such as pride and
shame: subjects were asked to recount experiences that had either made them proud,
or ashamed. They did so while simultaneously moving marbles from a lower tray to a
higher tray or vice versa. Lakoff and Johnson (1980, 1999) argue that positive experi-
ences are metaphorically conceptualized as being up, while negative experiences are
experienced as being down. Casasanto and Dijkstra found that the speed and efficiency
of the autobiographical retelling was influenced by whether the direction of the
marble movements was congruent with the autobiographical memory: upward for
pride, downward for shame. This provides compelling evidence that even abstract
language appears to involve automatic activation of sensorimotor simulations in the
brain: we understand what the words pride and shame mean, in part, by virtue of the
264 Vyvyan Evans

upward and downward trajectories that metaphorically structure them being activated
in the brain.
More generally, an important conclusion from this discussion is the following. The
traditional distinction between perception and cognition—an artifact of the earlier
distinction between body and mind arising from the seventeenth-century philosophi-
cal underpinnings of psychology—may be too strong (for discussion see Barsalou
1999; Bergen 2012; Prinz 2002). Representations that arise in language use and com-
prehension are grounded in the same knowledge that is used in processing our experi-
ences of the world around us. The distinction between perception and cognition, at
the very least, may not be as clear-cut as some cognitive scientists have claimed. Talmy
(2000), one of the pioneering linguists who first saw that language encodes embod-
ied concepts, argued for a unified category, which he termed ception; Talmy sought to
emphasize the continuity, rather than separation, between perception and conception
(or cognition).

10.4 Conceptual Structure versus Semantic Structure

In light of the foregoing, the conclusion I draw is this: language and body-based repre-
sentations would appear, together, to co-conspire in the integration process that gives
rise to meaning. The question is how, a question I begin to address in this section.
From an evolutionary perspective, the perceptual and interoceptive representations
in the conceptual system must have preceded language. The conceptual system allows
us, and many other species, to have available for reactivation the body-based repre-
sentations that arise from our interaction in our socio-physical world of experience.
Humans are not alone in possessing conceptual systems and, presumably, body-based
representations in those conceptual systems (Barsalou 2005; Hurford 2007, 2012). A
conceptual system enables an organism to represent the world it encounters, to store
experiences, and hence to respond to new experiences as a consequence. A conceptual
system is what enables us and other species to be able to tell friend from foe, com-
petitor from potential sexual mate, and to act and interact in situationally appropriate
ways. Our repository of concepts facilitates thought, categorization of entities in the
world, and our action and interaction with, in, and through the spatiotemporal envi-
ronment we inhabit.
But complex thoughts, actions, and so on require that our concepts can be com-
bined compositionally in order to form complex ideas. While many other species have
conceptual systems, humans appear to be unique in having language. And the range
and complexity of human thought appear to far exceed those of any other species. As
Bertrand Russell pithily observed, “No matter how eloquently a dog can bark, it cannot
tell you that its parents were poor but honest.” An obvious implication, then, is that
language may provide, in part at least, a means of harnessing our conceptual system,
What’s in a Concept? 265

of releasing its potential—a conclusion that has been reached by a number of leading
cognitive scientists (see, for example, Bergen 2012; Evans 2009; Mithen 1996,;and ref-
erences therein).
Barsalou (2005; Baraslou et al. 2008; see also Evans 2009) has suggested that the func-
tion of language is to provide an executive control function, operating over grounded
concepts in the conceptual system. And this view has much to commend it. The idea is
that language provides the framework that facilitates the composition of concepts for
purposes of communication. This is achieved by language consisting of a grammatical
system, with words and constructions cuing activations of specific body-based states in
the brain (see Bergen 2012, chapter 5). Their integration gives rise to complex simula-
tions, which is the stuff of thought. On this account, language provides added value.
It allows us to control and manipulate the conceptual system, which, after all, must
have originally evolved for more rudimentary functions, such as object recognition
and classification. Under the control of language, we can make use of body-based (not
exclusively sensorimotor) concepts in order to develop abstract thought. As Barsalou
et al. (2008) explain:

Adding language increased the ability of the simulation [=conceptual] system to represent
non-present situations (past, future, counterfactual). Adding language increased the ability to
construct simulations compositionally. Adding language increased the ability to coordinate simu-
lations between agents, yielding more powerful forms of social organisation. (274)

However, if the function of language is to index or activate body-based concepts, we

might reasonably ask what language is bringing to the table. Do words have meanings
in their own right, independent of the perceptual and interoceptive representations
they point to?
Some embodied mind researchers have denied that language contributes to mean-
ing per se. Representatives of this position argue that language has no semantic proper-
ties of its own, independent of the simulations produced by grounded concepts in the
conceptual system (see Glenberg and Robertson 1999; Barsalou et al. 2008).
One reason for thinking this is that decontextualized words on their own do not
trigger simulations. In contrast, analog representations such as pictures and images do
(Lindemann et al. 2006). For instance, being exposed to the word cat, unless embedded
in a sentence, will not normally, on its own, give rise to a particularly rich conceptu-
alization. In contrast, a picture of a cat, which is analog—it iconically represents our
visual experience of a cat—gives rise to a simulation.
Another line of evidence relates to gist. Upon hearing a story, subjects appear to store
the gist of the narrative but not the form that it was told in—subjects can recount the
story, but often use quite different words to do so. This takes place after about twenty
seconds, suggesting that while the language used to convey the story erodes, the story
itself is represented in a nonlinguistic form, a complex representation, a simulation, in
266 Vyvyan Evans

the conceptual system (Bransford and Franks 1971). This suggests that once a simula-
tion has been achieved, language is largely irrelevant for, and hence independent of,
the underlying meaning (Barsalou et al. 2008).
Barsalou and colleagues conclude from this that language processing is not very
deep, in terms of the semantic representations it can evoke on its own. The role of
language is to provide structure, which aids the assembly of perceptual states in the
construction of meaning. In other words, simulations arise from activation of non-
linguistic concepts. And it is these simulations that linguistic forms provide direct
access to. According to Barsalou (e.g., Barsalou et al. 2008), language provides a level,
essentially, of formal (sound or signed) representation, but no semantic content. The
forms then hook up with perceptual and interoceptive states, thereby facilitating
However, this view is likely to be too strong. First, if language has no independent
semantic content, then presumably we can’t use language to evoke ideas we haven’t
yet experienced—because the brain states don’t yet exist for the experiences. Yet, we
can use language to evoke experiences we haven’t yet experienced (Taylor and Zwaan
2009; Vigliocco et al. 2009). The experiences evoked via language, in the absence of a
fully “immersed” experience, such as seeing or enacting the experience, is somewhat
attenuated and full of gaps. Nevertheless, language can facilitate an approximation.
By way of illustration, consider the lutz jump. Readers largely ignorant of ice-skating
will also be ignorant of what this move entails. Now read the following definition of
the lutz jump:

A jump in figure skating in which the skater takes off from the back outer edge of one skate and
makes one full rotation before landing on the back outer edge of the other skate

Having read this, readers will have a rough idea of what the lutz jump is. They will
understand it is a move in ice-skating, which involves a particular kind of footwear
on a particular kind of surface. Moreover, readers will be able to close their eyes and
rehearse a move in which an ice-skater takes off, performs one rotation, and lands. To
be sure, many of the details will be unclear. If readers were then to look up lutz jump
on YouTube, they would be able to watch clips of the lutz jump being performed.
And this illustrates a point I will return to later: veridical experience, the experience of
seeing, acting, and interacting, gives rise to body-based representations that are analog
in nature. Having seen the lutz jump, readers can, thereafter, picture it in their mind’s
eye. Language, in contrast, doesn’t work like that, I contend. The representations are
more sketchy, more partial; they are not analog at all.
More troublesome for Barsalou, and for others who seek to exclude semantic content
from language, is the following: although simulations arise automatically in response
to language use, they are not necessary for language to be successfully used. Patients
with Parkinson’s disease who display difficulty in carrying out motor movements,
What’s in a Concept? 267

suggesting their motor representations are damaged, are still able to use and understand,
more or less, corresponding action verbs (Boulenger et al. 2008). Likewise, patients with
motor neuron disease are still able to process action verbs, albeit suboptimally (Bak
et al. 2001). Indeed, this was the one of the objections to an embodied approach to
concepts raised by Mahon and Caramazza (2008), discussed earlier.
Taylor and Zwaan (2009) account for this by proposing what they call the fault
tolerance hypothesis. This makes the following claim: humans construct their concep-
tual representations from various sources, including language. Moreover, these may be
incomplete. For instance, a novice skier doesn’t have the motor routines necessary for
skiing; an unconventional ski instructor might ask the novice skier to imagine being a
waiter, with a tray held aloft, weaving through a busy Parisian café, in order to simulate
the type of body posture and motor routines required when on skis.2 The point is that
evoking such a simulation, via language, while not the same as the embodied experi-
ence of skiing, facilitates the learning of the requisite motor routines that, in time, will
lead the novice to becoming an accomplished skier.
The third problem is this. The grammatical subsystem appears to encode semantic
content—albeit schematic content—independently of the conceptual system (Evans
2009; Evans and Green 2006; Talmy 2000; see also Bergen 2012, chapter 5). To illus-
trate, if we exclude the semantic content associated with open-class elements such as
nouns, verbs, adjectives, and adverbs, we are left with a type of schematic represen-
tation that is not straightforwardly imageable, or perceptual. In short, the represen-
tations associated with grammatical structure, so-called closed-class elements, appear
not to relate, in a straightforward way, with perceptual representations. And yet, such
representations appear to be meaningful. For instance, the distinction between the
definite article the and the indefinite article a is one of specificity. But it is not clear
what the might relate to in the conceptual system: although we can visualize open-
class lexical items, such as chair or tree, and simulate the feelings and scenes associated
with more abstract nouns such as love and war, we can’t simulate whatever it is that the
corresponds to, for instance.
To make this point more explicitly, consider the following:

(3) Those boys are painting my railings

In this example, if we strip away the open-class elements we are left with the
closed-class elements in bold—the bound morphemes (-ing and -s), and the closed-
class free morphemes those, are, and my. Moreover, the state of being a noun, which
is to say, the schematic category noun, and the schematic category verb (although not
specific exemplars of nouns, exemplified by boy, railing, and paint) are also closed-class

2. This example is due to Fauconnier and Turner (2002).

268 Vyvyan Evans

The composite meaning of all these closed-class elements in (3) is as follows: Those
somethings are somethinging my somethings. This can be glossed as follows: “More than
one entity close to the speaker is presently in the process of doing something to more
than one entity belonging to the speaker.” This actually provides quite a lot of mean-
ing. That said, this semantic representation for the closed-class entities is schematic.
We don’t have the details of the scene: we don’t know what the entities in question are,
nor do we know what is being done by the agent to the patient.
Nevertheless, this illustration reveals the following: there appears to be a type of
semantic representation that, arguably, is unique to the linguistic system. Moreover,
this representation provides information relating to how a simulation should be con-
structed (see Bergen 2012 for a related point). After all, the grammatical organization of
the sentence entails that the first entity is the agent and the second entity the patient:
the first entity is performing an action that affects the second entity. This level of
semantic representation derives exclusively from language, rather than from concep-
tual structure, providing an instruction as to the relative significance, and the rela-
tion that holds, between these two entities. In short, closed-class elements, and the
grammatical configurations in which they reside—which are themselves closed-class
elements—involve semantic content, albeit of a highly schematic sort (Evans 2009;
Goldberg 1995, 2006; Talmy 2000).
There is one further difficulty with assuming that language has no semantic content,
independent of conceptual structure. We now know that language appears to directly
influence perception. In a language such as Greek, for instance, there are distinct words
for light blue (ghalazio) and dark blue (ble). This contrasts with English, where there is
a single word: blue. Neuroscientists have shown that whether one is a native speaker
of Greek or of English influences how we perceive blueness. Using event-related poten-
tial (ERP) methodology, Thierry and colleagues (2009) found that the brain activity
of Greek speakers diverged when they perceived the different shades of blue. In con-
trast, English speakers exhibited no divergence in brain activity across the blue shades.
The conclusion that emerges from this is that there is clear relationship between a
linguistic distinction in a speaker’s native language—Greek divides blue color space
while English doesn’t—and the low-level, automatic perception of color, as measured
by brain activity at the onset of preattentive awareness, before subjects become con-
scious of the color they are perceiving. For present purposes, the relevance of this
finding is that it provides direct evidence that a parametric distinction imposed by a
language—dark versus light color—modulates nonlinguistic perceptual categorization
in visual experience. This could not be the case if language had no semantic content
independent of the conceptual system.
Finally, language appears to induce the illusion of semantic unity. This is an effect of
language rather than of the conceptual system. For instance, the word time in English
encodes a range of different, albeit related, concepts (Evans 2004):
What’s in a Concept? 269

(4a) The time for a decision has arrived.

(4b) The relationship lasted a long time.

(4c) Time flows on forever.

(4d) The time is approaching midnight.

In these sets of examples, all involving the lexical item time, a different reading
is obtained. In (4a), a discrete temporal point or moment is designated, without refer-
ence to its duration. The example in (4b) provides a reading relating to what might
be described as “magnitude of duration.” In the sentence in (4c), time prompts for
an entity that is infinite and hence eternal. Thus, in (4c) the reading relates to an
entity that is unbounded. Finally, the example in (4d) relates to a measurement sys-
tem, and specifically a point that could be paraphrased as “The hour is approaching
Although English has one word for a range of (arguably) quite distinct concepts,
other languages do not have a single word that covers all this semantic territory. For
instance, recent research on the Amazonian language Amondawa reveals that there is
no equivalent of the English word “time” in that language (Sinha et al. 2011). To give
another example of typologically distinct languages, Inuit languages also don’t have
a single lexeme for time (Michael Fortescue, pers. comm.). Moreover, even genetically
related languages use distinct lexical items to describe the semantic territory covered by
the single lexical form time in English.
French is a good example of this:

(5) C’est l’heure de manger

“It’s time to eat”

While the lexical form heure (hour) is used to describe the moment sense of time,
equivalent to the English example in (4a); some of the other senses for English time are
covered by the form temps (time). What this illustrates is that word forms can provide
an illusion of semantic unity (Evans 2009) and give rise to the myth that time, by way
of example, relates to a homogenous set of experiences. This is, I suggest, an effect of
language, rather than nonlinguistic knowledge, which remains broadly similar across
English and French. In short, other languages don’t group the same semantic terri-
tory with a single lexeme. Still others separate out across distinct lexemes. In the final
analysis, it appears that semantic unity is an illusion, an artifact of linguistic organiza-
tion and use. This provides compelling evidence that language brings with it its own
semantic contribution, independent of the rich and detailed knowledge representation
of the nonlinguistic conceptual (or simulation) system.
In sum, it is difficult to maintain the view held by some eminent embodied/
grounded cognition researchers that semantic structure in language equals conceptual
270 Vyvyan Evans

structure. There appear to be two distinct types of representations. I now turn to a con-
sideration of what these might be.

10.5 Parametric versus Analog Concepts

In this section, I consider the distinction between representations in the linguistic

system, and those that inhere in the conceptual (or simulation) system. But let’s
first consider what it means to say that language activates sensorimotor states. From
the present perspective, the idea is that words are in fact cues that index or point to
body-based states processed and stored by the brain (Evans 2009, 2013; Glenberg and
Robertson 1999; Fischer and Zwaan 2009. To illustrate, consider the use of red in the
following example sentences (adapted from Zwaan 2004):

(6a) The schoolteacher scrawled in red ink all over the pupil’s homework book.

(6b) The red squirrel is in danger of extinction in the British Isles

In the first example, the use of red evokes a bright, vivid red. In the second, a
dun or browny red is most likely evoked. This illustrates the following: The meaning
of red is not, in any sense, there in the word (although I nuance this view below).
Rather, words cue perceptual and interoceptive states stored in the brain. And these
body-based states are reactivated during language use. Put another way, the word
form red gives rise to distinct simulations for different hues of red. These simulations
arise as a consequence of reactivating stored experience types. These reactivated expe-
riences we might refer to as analog concepts—concepts that are directly grounded in
the experiences that give rise to them. How then does semantic structure (in lan-
guage) differ from this level of conceptual structure—which is to say, from analog
To illustrate, I consider the use of the adjective red, and the noun redness, in the
following examples, adapted from a skin-care product advertisement:

(7a) Treat redness with Clinique urgent relief cream.

(7b) Treat red skin with Clinique urgent relief cream.

Both words, red and redness, relate to the same perceptual state, presumably. But the
words package or serve to construe the content in a different way, giving rise to distinct
simulations. In the example in (7a), redness gives rise to an interpretation relating to a
skin condition. In the second, (7b), red refers more straightforwardly to an unwanted
property of the skin.
The different interpretations arising from these sentences are not due to a different
hue being indexed. Rather, the words (noun versus adjective) modulate the percep-
tual hue in a slightly different way, giving rise to slightly distinct simulations: “skin
What’s in a Concept? 271

Analogue concept:

Parametric concept:
Property Thing

Word forms:
red redness

Figure 10.3 (plate 10)

Analog and parametric concepts.

condition” versus “discoloration of skin” interpretations. In other words, the gram-

matical distinction between the adjective (red) and noun (redness) appears to relate to
a semantic distinction between the notion of property versus thing. The words red and
redness, while indexing the same (or similar) perceptual state, also encode schematic
concepts: PROPERTY versus THING (cf. Langacker 2008).
But unlike the body-based perceptual state—the hue of red—which is analog,
PROPERTY and THING are highly schematic notions. They are what I refer to as parametric

concepts. Unlike the perceptual experience of redness, with comes to mind when we
variously imagine a fire engine, a Royal Mail post box (ubiquitous in the UK), henna,
fire, the Chinese flag, or superman’s cape, parametric concepts are not like veridical
embodied experiences. There is nothing about the (parametric) concepts PROPERTY or
THING that is akin to the perceptual experience of redness (an analog concept). Param-

eters are abstracted from embodied (perceptual and interoceptive) states, filtering out
all points of difference to leave highly image-schematic content: the parameter.3 The
word form red encodes the parameter PROPERTY, while redness encodes the parameter
THING. This is another way of saying that red is an adjective—it describes a property of a

thing—while redness is a noun—it describes a property that is objectified in some way

and established as being identifiable, in principle, in its own right, independent of
other entities in world. Figure 10.3 (plate 10) captures the relationship between a word
form and its parametric and analog concepts.

3. Cf. the related notion of image schema developed in the work of Johnson (1987).
272 Vyvyan Evans

My claim, then, is this. There is a distinction between analog concepts on the

one hand and parametric concepts on the other. The former relate to nonlinguistic
concept types that, in evolutionary terms, had to precede the existence of language.
Parametric concepts constitute a species of concept that arose as a consequence of
the emergence of language. They provide a level of schematic representation directly
encoded by language: parametric concepts guide how analog concepts are activated
and, consequently, how simulations are constructed in the service of linguistically
mediated meaning construction. For instance, the forms red and redness both index
the same perceptual state(s). But they parcellate the conceptual content in a differ-
ent way, giving rise to distinct simulations: redness = condition; red = (unwanted)
property of skin. The schematic parametric concepts, which is to say, that part
of meaning that is native to language, relates to THING versus PROPERTY. Parametric
concepts are language-specific affordances, rather than affordances of the simula-
tion system.
Related proposals have been put forward by Bergen (2012) and Taylor and Zwaan
(e.g., 2008, 2009). Taylor and Zwaan have captured this view in terms of what they dub
the linguistic focus hypothesis. They argue that during language understanding, motor
representations are activated that are under the governance of linguistic constructions.
These serve to differentially direct focus on the referential world. Bergen’s findings are
consonant with this hypothesis. In one set of behavioral experiments, Bergen (2012)
found that the grammatical subject, for instance, the use of I versus you, influences
the perspective from which a language user perceives a scene. Bergen explains this as

Grammatical person seems to modulate the perspective you adopt when performing embodied
simulation. This isn’t to say that every time you hear you, you think about yourself as a partici-
pant in simulated actions. But it does mean that the grammatical person in a sentence pushes
you toward being more likely to adopt one perspective or another. What’s interesting about this
is that in this case, grammar appears not to be telling you what to simulate, but rather, how to
simulate—what perspective to simulate the event from. Instead of acting as the script in this case,
grammar is acting as the director. (114)

In the light of this discussion, what then is the function of language and, specifically,
parametric concepts in embodied cognition? My answer is that parametric concepts,
encoded by language, guide the formation of complex simulations for purposes of
(linguistically mediated) communication. Parametric concepts guide the parcellation
(focal adjustments, in Langacker’s 2008 terms) of analog (i.e., perceptual and interocep-
tive) concepts, in the construction of simulations. Parametric concepts encode sche-
matic, or digitized, content. Content of this sort is abstracted from analog, or perceptual
(and interoceptive) representations. Hence, the parameters THING versus PROPERTY are
schemas drawn from embodied experience.
What’s in a Concept? 273

Let’s now examine a more complex example of a parametric concept. Consider the
ditransitive construction (Goldberg 1995), as exemplified by the following:

(8) John baked Mary a cake.

Goldberg argues that this example is sanctioned by a sentence-level construction

that encodes the schematic semantics in (9):


Goldberg’s point is that the “cause to receive” meaning in (9) arises not from the
semantics of bake, which is canonically a transitive (rather than a ditransitive) verb,
but from the larger construction in which it is embedded. And there is behavioral
evidence to support such a contention. Kaschak and Glenberg (2000) reported a study
in which they showed sentences to participants using the novel verb to crutch. Some
sentences employed the ditransitive construction, as in (10a), while others placed
the novel verb in the transitive construction as in (10b). They then asked subjects to
say which of the sentences were consistent with two inference statements, given below
in (11):

(10a) Lyn crutched Tom her apple

(10b) Lyn crutched her apple

(11a) Tom got the apple

(11b) Lyn acted on her apple

The sentence in (11a) provides an inference of transfer of possession. In contrast, the

inference arising from (11b) is to act on. Because the verb is novel, it has no inherent
semantics associated with it. Hence, if the sentence possesses inherent semantics inde-
pendently to the verb, as claimed by Goldberg, then we would expect the inference in
(11a) to be judged as compatible with sentence (10a), and the inference in (11b) to be
compatible with the sentence in (10b). And this is indeed what Kaschak and Glenberg
found. In short, the syntactic sentence-level template appears to have a schematic rep-
resentation associated with it—a complex parametric concept in present terms—which
is represented in (9). Parametric concepts of this sort guide or modulate how analog
concepts are parcellated, giving rise to a simulation. And a complex parametric concept
such as (9) does this, in principle, in the same way as parametric concepts associated
with single lexical items such as red and redness.
In essence, it turns out that working out what a concept is, is not a straightforward
matter at all. Body-based representations, stored in different brain regions, form the
basis of a species of concepts: analog concepts. Concepts of this kind are grounded
in the veridical (perceptual) and phenomenological (interoceptive) experience types
from which they arise. But a second species of concept, parametric concepts, appears
274 Vyvyan Evans

Table 10.2
Parametric versus analog concepts

Parametric concepts Analog concepts

Specific to language Specific to the conceptual system

Parametric (abstracted from embodied Analog (albeit attenuated)
states, filtering out all points of difference representations of body-based states
to leave highly schematic properties or Arise directly from perceptual (conscious)
parameters) experience and reside in the same neural
Underpinnings for all linguistic units system(s) as body-based states
(where a linguistic unit is a form or Reactivated or simulated (by language,
parametric content unit of any complexity) imagination, etc.) and can be combined
to form complex and novel simulations

to be directly encoded by language. Concepts of this kind are far more schematic:
they encode parameters—THING versus PROPERTY, DARK COLOR versus LIGHT COLOR. They
are abstracted from embodied experience but are not rich analog representations.
Moreover, the parametric concepts appear to be deployed to modulate the analog
concepts in giving rise to a representation known as a simulation: a complex repre-
sentation that is constructed in the process of speaking and thinking. This simulation
expresses an idea that language is instrumental in facilitating. Table 10.2 summarizes
the distinction between parametric and analog concepts.

10.6 Access Semantics

Having distinguished between analog and parametric concepts, we now require

an account of the respective contributions of parametric and analog concepts to
the meaning-construction process. In other words, we need an account of how
parametric concepts access analog concepts. To do this, in this section I develop a
theory of access semantics. An access semantics accounts for how semantic structure
(made up of parametric concepts) interfaces with, and thereby activates, the requi-
site aspects of nonlinguistic knowledge representation—that is, conceptual structure
(made up of analog concepts)—which inheres in the conceptual system, giving rise to
a simulation.

10.6.1 LCCM Theory

In Evans (2009, 2013) I have developed a theoretical account of lexical representa-
tion and semantic composition dubbed the theory of lexical concepts and cognitive
models, or LCCM theory for short. LCCM theory is a theory of access semantics. The
claim at its heart is enshrined in the distinction between its two foundational theoreti-
cal constructs—the lexical concept and the cognitive model: there is a qualitative distinc-
tion between these theoretical constructs, which are central to meaning construction.
What’s in a Concept? 275

This distinction relates, ultimately, to the bifurcation between analog and parametric
concepts, which respectively structure cognitive models and lexical concepts.
In keeping with the thrust of the argument developed in the foregoing, LCCM the-
ory assumes the linguistic system emerged, in evolutionary terms, much later than the
earlier conceptual system. The utility of a linguistic system, on this account, is that
it provides an executive control mechanism facilitating the deployment of concep-
tual representations in service of linguistically mediated meaning construction. Hence,
semantic representations in the two systems are of a qualitatively distinct kind. I model
semantic structure—the primary representational substrate of the linguistic system—in
terms of the theoretical construct of the lexical concept. A lexical concept is a compo-
nent of linguistic knowledge—the semantic pole of a symbolic unit (in, e.g., Langacker’s
1987 terms)—which encodes a bundle of various types of highly schematic linguistic
content (see Evans 2006, 2009, 2013). In particular, linguistic content includes informa-
tion relating to the selectional tendencies associated with a given lexical concept—the
range of semantic and grammatical correlates of a given lexical concept (see Evans
2006, 2009). Hence, lexical concepts are parametric concepts.
While lexical concepts encode highly schematic linguistic content, a subset—those
associated with open-class forms—are connected, and hence facilitate access, to the
conceptual system. Lexical concepts of this type are termed open-class lexical concepts.4
Such lexical concepts are typically associated with multiple areas in the conceptual
system, referred to as association areas.
The range of association areas to which a given lexical concept facilitates access is
termed an access site. LCCM theory assumes that the access site for a given open-class
lexical concept is unique. As lexical concepts facilitate access to a potentially large num-
ber of association areas in the conceptual system, any given open-class lexical concept,
in principle, facilitates access to a large semantic potential. However, only a small subset
of this semantic potential is typically activated in interpretation of a given utterance.
Although the linguistic system evolved to harness the representational power of the
conceptual system for purposes of communication, the human conceptual system, at
least in very broad outline, is continuous with that of other primates (Barsalou 2005;
Evans 2013, especially chapter 2, 2014), and shows a range of broad similarities with
that of other species (Hurford 2007). In contrast to the linguistic system, the concep-
tual system evolved primarily to facilitate functions such as perception, categorization,
inference, choice, and action, rather than communication. In LCCM theory, conceptual
structure—the semantic representational substrate of the conceptual system—is mod-
eled by the theoretical construct of the cognitive model. A cognitive model is a coher-
ent body of multimodal knowledge grounded in the brain’s modal systems, and derives
from the full range of experience types processed by the brain, including sensorimotor

4. See Evans (2009) for the rationale for this position.

276 Vyvyan Evans

Lexical representation

Conceptual system Linguistic system

Cognitive model Symbolic unit


Figure 10.4
Relationship between a lexical concept and a cognitive model.

experience, proprioception, and subjective experience, including affect.5 Hence, cogni-

tive models are analog in nature, and as such constitute analog concepts.
The conceptual content encoded as cognitive models can become reactivated during
the simulation process. Simulation is a general-purpose computation performed by the
brain to implement the range of activities that subserve a fully functional conceptual
system. Such activities include conceptualization, inferring, choice, categorization, and
the formation of ad hoc categories.6
In LCCM theory, simulations are effected by a subset of lexical concepts—open-class
lexical concepts—facilitating access to the conceptual system via a number of associa-
tion areas (see figure 10.4). Each association area corresponds to a (part of a) cognitive
model, as captured in figure 10.4. The range of association areas to which an open-class
lexical concept corresponds makes up its access site.

10.6.2 Cognitive Model Profile

An important construct in LCCM theory, and one that is essential to providing an
account of meaning construction, is that of the cognitive model profile. As an open-
class lexical concept—a noun, verb, adjective, or adverb—facilitates access to numerous

5. The term cognitive model is used elsewhere in cognitive science, for instance in terms of compu-
tational modeling (e.g., in John Anderson’s ACT-R theory of cognition), and is widespread in this
other sense. My use of the term is not being deployed in the same way.
6. For discussion and findings relating to the multimodal nature of conceptual representations
and the role of simulation in drawing on such representations in facilitating conceptual function,
see, for instance, Barsalou (1999, 2008), Glenberg (1997), Gallese and Lakoff (2005), and refer-
ences therein.
What’s in a Concept? 277

Constitutional Head of
system state

National Political
sports system

Geographical Nation Holiday

landmass state destination


Figure 10.5
Partial cognitive model profile for the lexical concept [FRANCE].

association areas within the conceptual system, it facilitates access to numerous cog-
nitive models. Moreover, the cognitive models to which a lexical concept facilitates
access are themselves connected to other cognitive models. The range of cognitive
models to which a given lexical concept facilitates direct access, and the range of addi-
tional cognitive models to which it therefore facilitates indirect access, are collectively
termed its cognitive model profile.
To illustrate, consider the cognitive model profile for the lexical concept I gloss as
[FRANCE], associated with the form France. A partial cognitive model profile for [FRANCE]
is represented in figure 10.5.
Figure 10.5 is an attempt to capture the sort of knowledge that language users must
presumably have access to when speaking and thinking about France. As illustrated by
figure 10.5, the lexical concept [FRANCE] provides access to a potentially large number
of cognitive models.7 Because each cognitive model consists of a complex and struc-
tured body of knowledge, which, in turn, provides access to other sorts of knowledge,
LCCM theory distinguishes between cognitive models that are directly accessed via the
lexical concept—primary cognitive models—and those that form substructures of directly

7. The bracket notation used here, [FRANCE], represents the linguistic content that is encoded by
the vehicle “France.”
278 Vyvyan Evans

accessed models—secondary cognitive models. These secondary cognitive models are

indirectly accessed via the lexical concept.
The lexical concept [FRANCE] affords access to a number of primary cognitive models,
which make up the primary cognitive model profile for [FRANCE]. These are hypothesized
a cognitive model represents a coherent body of complex information, multimodal
information, gleaned through sense perception, interoceptive experience, and prop-
ositional information achieved via cultural learning, language, and other channels.
Each of these cognitive models provides access to further cognitive models. In figure
10.5, a flavor of this is given by virtue of the various secondary cognitive models that
are accessed via the NATION-STATE cognitive model—the secondary cognitive model profile.
These include NATIONAL SPORTS, POLITICAL SYSTEM, and CUISINE. For instance, we may know
that in France, the French engage in national sports of particular types, for instance,
football, rugby, and so on, rather than others: the French don’t typically engage in
American football, ice hockey, cricket, and so on. We may also know that as a sporting
nation, they take part in international sports competitions of various kinds, including
the FIFA World Cup, the Six Nations rugby competition, the Rugby World Cup, the
Olympics, and so on.
That is, we may have access to a large body of knowledge concerning the sorts
of sports French people engage in. We may also have some knowledge of the fund-
ing structures and social and economic conditions and constraints that apply to these
sports in France as well as France’s international standing with respect to these particu-
lar sports, and further knowledge about the sports themselves, including the rules that
govern their practice and so forth. This knowledge is derived from a large number of
sources, including direct experience and cultural transmission (including language).
With respect to the secondary cognitive model of POLITICAL SYSTEM, figure 10.5
illustrates a sample of further secondary cognitive models that are accessed via this
cognitive model. In other words, each secondary cognitive model has further (second-
ary) cognitive models to which it provides access. For instance (FRENCH) ELECTORATE is a
cognitive model accessed via the cognitive model (FRENCH) POLITICAL SYSTEM. In turn the
cognitive model (FRENCH) POLITICAL SYSTEM is accessed via the cognitive model NATION-STATE.
Accordingly, NATION-STATE is a primary cognitive model, while ELECTORATE and POLITICAL
SYSTEM are secondary cognitive models.

Finally, it is worth highlighting a point that has been implicit in the foregoing.
LCCM theory assumes that cognitive models involve content at varying degrees of
abstractness, and of different types. For instance, being able to simulate the view from
the top of Mont Saint-Michel on a midsummer evening, while on vacation in France,
involves processing sensorimotor aspects of a scene, as well as knowledge of the affec-
tive experiences that accompany this view—pleasure, joy, awe, and so on. In addition,
such information is subject to abstraction, giving rise to schematic categories, such as
What’s in a Concept? 279

Sue type
buys type Petrol
Driver Fuel
type Diesel
Mike type
operates flows
aspect type 4 cylinder
Engine type
CAR aspect 6 cylinder

type 8 cylinder

type Manual
type Automatic

type Steel
type Alloy

Figure 10.6
Knowledge structure for CAR concept.

“beautiful places I have seen,” which may come to form part of the cognitive model
profile for [FRANCE], but may also become linked to other cognitive model profiles. More-
over, cognitive models may contain information relating to entities, both individuals
and types, as well as relational information. For instance, a cognitive model profile for
[CAR] will include knowledge of various sorts, including, in various levels of detail con-
tingent on the individual’s knowledge base, the component parts of a car and the rela-
tions between them, as illustrated schematically in figure 10.6. Moreover, as language
interfaces with the conceptual system via access sites, cultural knowledge mediated by
language can come to form part of the representations associated with cognitive mod-
els. In short, LCCM theory holds that analog (body-based) content is supplemented by
propositional information derived from linguistically mediated content, thus fleshing
out the representations in the conceptual system.

10.6.3 Activation of Cognitive Models in Meaning Construction

The way meaning construction proceeds is by virtue of integration of parametric
concepts, which give rise to a linguistic context that facilitates activation of aspects
280 Vyvyan Evans

of cognitive model profiles. The outcome, then, of language understanding is activa-

tion of nonlinguistic cognitive models, guided and constrained by linguistic context.
Consider the following linguistic examples:

(12a) France is a country of outstanding natural beauty.

(12b) France is one of the leading nations in the European Union.

(12c) France beat New Zealand in the 2007 Rugby World Cup.

(12d) France voted against the EU constitution in the 2005 referendum.

In each of these examples, the semantic contribution associated with the form
France is slightly distinct. That is, the semantic contribution provided by France varies
across these distinct utterances. The key insight of LCCM theory is that the reason for
this variation is due to differential activation of nonlinguistic knowledge structures,
the cognitive model profile, to which the lexical concept associated with France affords
The informational characterization associated with [FRANCE] in each of these examples
concerns France as a geographical landmass in (12a), France as a political entity, a
nation-state, in (12b), the fifteen players who make up the French rugby team in (12c),
and in (12d), that proportion of the French electorate who voted “non” when pre-
sented, in a recent referendum, with the proposal to endorse a constitution for the
European Union. In order to provide these distinct interpretations, this lexical concept
must serve as an access site for a cognitive model profile that, at the very least, includes
the sort of information indicated in figure 10.5.
The differential interpretations associated with the examples in (12) arise as follows.
In (12a) the interpretation associated with the form France, which relates to a par-
ticular geographical region, derives from activation of the GEOGRAPHICAL LANDMASS cogni-
tive model. That is, individual language users have knowledge relating to the physical
aspects of France, including its terrain and its geographical location. In this example,
the utterance context activates this part of the cognitive model profile accessed by
the lexical concept [FRANCE]. In the second example, the utterance context activates
a different part of the cognitive model profile to which the lexical concept [FRANCE]
affords access. In this example, the informational characterization relates to the cog-
nitive model of France as a POLITICAL ENTITY. This is due to activation of the NATION-STATE
cognitive model. The use of France in the example in (12c) relates to the group of fifteen
French individuals who play as a team and thereby represent the French nation on
the rugby field. In the example in (12d), the form France relates not to a geographical
landmass, political entity, nation-state, nor group of fifteen rugby players who happen
to be representing the entire population of France. Rather, it relates to that portion
of the French electorate that voted against ratification of the EU constitution in a ref-
What’s in a Concept? 281

erendum held in 2005. Accordingly, what is activated here is the ELECTORATE cognitive
This last example provides an elegant illustration of the way in which activation of a
cognitive model provides a situated interpretation of a lexical concept by giving rise to
an access route through the semantic potential. In this example, interpretation requires
that an access route be established through the cognitive model profile accessed via the
lexical concept [FRANCE] in a way that is consistent with the lexical concepts associated
with the other linguistic forms and units in the utterance. The interpretation associated
with France in this example has to do with the French electorate, and specifically that
part of the French electorate that voted against ratification of the EU constitution. In
other words, [FRANCE] in this example achieves an informational characterization that is
facilitated by activating the cognitive models shown in bold in figure 10.7.

10.6.4 Matching
My final illustration speaks directly to the criticism leveled by Mahon and Caramazza
(2008) against embodied/grounded approaches to concepts. They argued that an
embodied account of concepts can’t deal with the variable properties evoked by beauti-
ful in these sorts of examples:

Constitutional Head of
system state

National Political
sports system

Geographical Nation Holiday

landmass state destination


Figure 10.7
Access route established by the interpretation of [FRANCE] in the utterance “France voted against
the EU constitution.”
282 Vyvyan Evans

(13a) a beautiful face

(13b) a beautiful sound

(13c) a beautiful idea

On the contrary, accounting for the variable sensorimotor and affective informa-
tion evoked by beautiful in these examples turns on the issue of meaning construction,
rather than (solely) on an account of knowledge representation. As such, it is guided
by the linguistic context.
The challenge, in the examples in (12), is to account for how the conceptual content
for beautiful, and face are integrated. This involves, in LCCM terms, a process of match-
ing across cognitive model profiles accessed by the relevant open-class lexical concepts.
Moreover, this matching is based on conceptual coherence across the cognitive model
profiles to ensure that the “correct” cognitive models become activated, leading to a
To begin to illustrate, consider a partial cognitive model profile for the open-class
lexical concept [BEAUTIFUL]—see figure 10.8. Primary cognitive models that are accessed
by [BEAUTIFUL] range from assessments relating to the receipt of or awareness of visual
pleasure, particularly physical appearance, often related to perceived sexual attractive-
ness, to the awareness of nonvisual but physical pleasure, such as aural pleasure, as in
the appreciation of music, or pleasure derived from touch, for instance. The lexical
concept [BEAUTIFUL] also affords access to a cognitive model having to do with nonphysi-
cal pleasure, which I gloss as AESTHETIC PLEASURE. This relates to an appreciation for such
things as art and literature from which pleasure are derived.
In the examples in (12), the linguistic context, a noun phrase (NP), gives rise to
the parametric concept [A NONSPECIFIC THING WITH A PARTICULAR ATTRIBUTE]. This schematic
semantic representation drives the matching process. That is, the two open-class lexical
concepts, for instance [BEAUTIFUL] and [FACE], afford access to their respective cognitive
model profiles. But they do so in a way that is consistent with the parcellation provided

Visual Non-visual Aesthetic

Pleasure physical pleasure pleasure


Figure 10.8
Partial cognitive model profile for the lexical concept [BEAUTIFUL].
What’s in a Concept? 283

by the parametric concept: whatever cognitive model [BEAUTIFUL] activates, this must be
interpretable as an attribute associated with face. And vice versa, the cognitive model
activated by [FACE] must be coherent with its property beautiful.
The process involved is thus one of matching, guided by the parametric concept.
As such, because a face is a physical entity that can be seen, conceptual coherence is
achieved when the VISUAL PLEASURE cognitive model is activated. In short, the language
constrains the matching process involved in activation of nonlinguistic concepts, in
service of linguistically mediated simulations.

10.7 Conclusion

I began this chapter by asking this question: Do linguistic units (e.g., words) have
semantic content independently of the human conceptual system? Some recent the-
ories of embodied (or grounded) cognition, for instance LASS theory, developed by
Barsalou and colleagues (2008), have tended to assume that the value of language is
in providing syntactic organization that facilitates the assembly of nonlinguistic con-
cepts, and hence simulations. On this view, language has no semantic contribution
independent of nonlinguistic concepts. In contrast, some psycholinguists (e.g., Taylor
and Zwaan 2008, 2009) and linguists (e.g., Bergen 2012) take the view that language
shapes the nature of simulations, by providing a level of focus on how the simulation
should be constructed. The implication of such a view, I suggest, is that language must
have a level of semantic content independent of nonlinguistic concepts. The distinc-
tion between the two, between semantic structure versus conceptual structure, I have
argued, can be operationalized in terms of the distinction between what I refer to as
parametric information (or concepts) and analog information (or concepts). Lexical
concepts, the concepts specific to the linguistic system, are parametric, providing a
level of schematic information. On this view, a lexical concept, made up of parameters,
is the semantic pole of a linguistic unit, where a linguistic unit is a symbolic assem-
bly of form and (schematic) meaning. In addition, lexical concepts facilitate access to
nonlinguistic concepts, which, in LCCM theory terms, are labeled cognitive models.
Cognitive models consist of analog information.
The parametric information that makes up lexical concepts can be associated with
single word forms, more complex idiomatic units, and even sentence-level construc-
tions. From this view, the parameters encoded in language actively shape, or parcel-
late, in my terms, the analog content to which a subset of lexical concepts (open-class
lexical concepts) facilitate access. For instance, the ditransitive construction, described
earlier, stipulates the relationship between the lexical concepts associated with the two
NPs in the construction. One entity associated with NP1 is designated as recipient, and
the other, NP2, as object of transfer. Although the open-class lexical concepts associ-
ated with these NPs access analog information, via the cognitive models that form
284 Vyvyan Evans

their access sites, the simulations that are constructed arise based on the parcellation
facilitated by the parametric content. In short, my argument is that there must be a
distinction between parametric concepts (encoded by language) and analog concepts,
associated with cognitive models, if linguistically mediated simulations are to arise in
the way claimed.
The consequence of the approach to knowledge representation and meaning con-
struction I develop is this. A grounded cognition approach to concepts has a means of
addressing criticisms that an account of abstract concepts, such as BEAUTIFUL, remains
intractable in this framework. On the contrary, with a descriptively adequate account
of knowledge representation—recognizing the distinction between analog and para-
metric concepts—and an account of meaning construction—the way in which linguis-
tic and simulation systems interface—such an account becomes conceivable, as I hope
to have demonstrated.
Moving forward, I see future research on conceptual structure as having three dis-
tinct goals. First, ongoing and future research needs to develop and refine empirically
verified accounts of knowledge representation. I have been assuming, based on current
knowledge, that cognitive models and cognitive model profiles are based on hierarchi-
cal knowledge structures, which are relational, are dynamically updated, and feature
conceptual distance between an access site—the cognitive models that interface with
the linguistic system. Evidence from linguistics suggests that some types of knowl-
edge indexed by words are more central to word meaning than others. The classic
distinction between denotation and connotation is one way in which this has been
operationalized (see Allan 1997). The so-called encyclopedic approach to representa-
tions underpinning words is another attempt to capture this (e.g., Langacker 1987).
Moreover, experimental psychology has offered abundant evidence that knowledge
is structured, ranging from classic work on typicality effects within prototype theory
(e.g., Rosch 1975, 1978; Rosch and Mervis 1975) to priming studies illustrating
relations between word meanings (e.g., Thompson-Schill, Kurtz, and Gabrieli 1998).
Psychophysical tasks have also demonstrated that knowledge is relational (see Barsalou
1999 for a review) and is dynamically updated (Elman 2009, 2011). Further research in
this view is projected under the aegis of LCCM theory, to develop an empirically robust
account of knowledge representation.
A second goal for future research must be to determine the processes that facili-
tate access to knowledge representation. Part of this investigation must involve the
compositional mechanisms that facilitate novel concept formation. For instance,
what makes it possible to produce ad hoc concepts such as things to take on vacation,
or things to remove from one’s house in the event of fire (Barsalou 1983). Equally,
how are novel blends produced, such as PET FISH—a nonprototypical fish (not gray),
and a nonprototypical pet (not furry)—and mythological creatures such as UNICORNS.
What’s in a Concept? 285

Fauconnier and Turner (2002; see also Turner 2014) have argued for an integrative
process termed blending that produces such imaginative feats. Hence, in addition to a
model of conceptual structure, we also presumably require the empirical study of the
conceptual integration processes that work on knowledge representation (see Coulson
2008 for a survey of preliminary work in this regard).
Third and finally, a future goal must surely be to better delineate and build on
the programmatic proposals above, for understanding the way in which the linguis-
tic system interfaces with the conceptual system in service of linguistically mediated
meaning construction. One way in which this might proceed is by empirical investi-
gation relating to the constructs of analog and parametric concepts. The claim is that
parametric concepts, devoid of a linguistic context, should not result in multimodal
brain activation, while analog concepts should. This prediction is something that, ulti-
mately, should be falsifiable. Other areas of research will no doubt need to focus on the
way in which linguistic constructions shape and modify simulations that arise. And
behavioral and ultimately brain-imaging research will need to be brought to bear on
this area.


Allan, K. 2007. The pragmatics of connotation. Journal of Pragmatics 39:1047–1057.

Bak, T. H., D. G. O’Donovan, J. H. Xuereb, S. Boniface, and J. R. Hodges. 2001. Selective impair-
ment of verb processing associated with pathological changes in Brodmann areas 44 and 45 in
the motor neuron disease-dementia-aphasia syndrome. Brain 124:103–120.

Barsalou, L. 1983. Ad hoc categories. Memory & Cognition 11 (3): 211–227.

Barsalou, L. 1999. Perceptual symbol systems. Behavioral and Brain Sciences 22:577–660.

Barsalou, L. 2005. Continuity of the conceptual system across species. Trends in Cognitive Sciences

Barsalou, L. 2008. Grounded cognition. Annual Review of Psychology 59:617–645.

Barsalou, L. W., A. Santos, W. K. Simmons, and C. D. Wilson. 2008. Language and simulation
in conceptual processing. In Symbols, Embodiment, and Meaning, ed. M. De Vega, A. Glenberg, and
A. Graesser, 245–283. Oxford: Oxford University Press.

Bergen, B. K. 2012. Louder than Words. New York: Basic Books.

Bonato, M., K. Priftis, R. Marenzi, C. Umiltà, and M. Zorzi. 2010. Increased attentional demands
impair contralesional space awareness following stroke. Neuropsychologia 48:3934–3940.

Boroditsky, L. 2000. Metaphoric structuring: Understanding time through spatial metaphors.

Cognition 75 (1): 1–28.
286 Vyvyan Evans

Boulenger, V., L. Mechtou, S. Thobois, E. Broussolle, M. Jeannerod, and T. A. Nazir. 2008. Word
processing in Parkinson’s disease is impaired for action verbs but not for concrete nouns. Neuro-
psychologia 46:743–756.

Bransford, J. D., and J. J. Franks. 1971. The abstraction of linguistic ideas. Cognitive Psychology

Buccino, G., L. Riggio, G. Melli, F. Binkofski, V. Gallese, and G. Rizzolatti. 2005. Listening to
action-related sentences modulates the activity of the motor system: A combined TMS and behav-
ioral study. Cognitive Brain Research 24:355–363.

Büchel, C., C. Price, and K. Friston. 1998. A multimodal language region in the ventral visual
pathway. Nature 394:274–277.

Bueti, D., and V. Walsh. 2009. The parietal cortex and the representation of time, space, number
and other magnitudes. Philosophical Transactions of the Royal Society of London: Series B, Biological
Sciences 364:1831–1840.

Casasanto, D., and L. Boroditsky. 2008. Time in the mind: Using space to think about time.
Cognition 106:579–593.

Casasanto, D., and K. Dijkstra. 2010. Motor action and emotional memory. Cognition 115 (1):

Chemero, A. 2009. Radical Embodied Cognitive Science. Cambridge, MA: MIT Press.

Cisek, P. 1999. Beyond the computer metaphor: Behaviour as interaction. Journal of Consciousness
Studies 6 (11–12): 125–142.

Clark, A. 1997. Being There: Putting Brain, Body and World Together Again. Cambridge, MA: MIT

Coulson, S. 2008. Metaphor comprehension and the brain. In R.W. Gibbs (ed.), Cambridge
Handbook of Metaphor and Thought, 177–196. Cambridge: Cambridge University Press.

Damasio, A. 1994. Descartes’ Error: Emotion, Reason and the Human Brain. New York: Penguin.

Dennett, D. 1969. Content and Consciousness. London: Routledge.

Dove, G. 2009. Beyond perceptual symbols: A call for representational pluralism. Cognition

Elman, J. L. 2009. On the meaning of words and dinosaur bones: Lexical knowledge without a
lexicon. Cognitive Science 33:1–36.

Elman, J. L. 2011. Lexical knowledge without a mental lexicon? Mental Lexicon 60:1–33.

Evans, V. 2004. The Structure of Time: Language, Meaning and Temporal Cognition. Amsterdam:
John Benjamins.

Evans, V. 2006. Lexical concepts, cognitive models and meaning-construction. Cognitive Linguis-
tics 17 (4): 491–534.
What’s in a Concept? 287

Evans, V. 2009. How Words Mean: Lexical Concepts, Cognitive Models and Meaning Construction.
Oxford: Oxford University Press.

Evans, V. 2010. Figurative language understanding in LCCM Theory. Cognitive Linguistics 21 (4):

Evans, V. 2013. Language and Time: A Cognitive Linguistics Perspective. Cambridge: Cambridge
University Press.

Evans, V. 2014. The Language Myth: Why Language is Not an Instinct. Cambridge: Cambridge
University Press.

Evans, V., and M. Green. 2006. Cognitive Linguistics: An Introduction. Edinburgh: Edinburgh
University Press.

Fauconnier, G., and M. Turner. 2002. The Way We Think: Conceptual Blending and the Mind’s
Hidden Complexities. New York: Basic Books.

Fischer, M. H., and R. A. Zwaan. 2009. Grounding Cognition in Perception and Action: A Special Issue
of the Quarterly Journal of Experimental Psychology. Hove: Psychology Press.

Fodor, J. 1975. The Language of Thought. Cambridge, MA: Harvard University Press.

Fodor, J. 1983. The Modularity of Mind. Cambridge, MA: MIT Press.

Fodor, J. 1998. Concepts: Where Cognitive Science Went Wrong. Oxford: Oxford University Press.

Fodor, J. 2008. LOT 2: The Language of Thought Revisited. Oxford: Oxford University Press.

Gallese, V., and G. Lakoff. 2005. The brain’s concepts: The role of the sensory-motor system in
reason and language. Cognitive Neuropsychology 22:455–479.

Glenberg, A. M. 1997. What memory is for. Behavioral and Brain Sciences 20:1–55.

Glenberg, A. M., and M. Kaschak. 2002. Grounding language in action. Psychonomic Bulletin &
Review 9 (3): 558–565.

Glenberg, A. M., and D. A. Robertson. 1999. Indexical understanding of instructions. Discourse

Processes 28 (1): 1–26.

Goldberg, A. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago:

University of Chicago Press.

Goldberg, A. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford
University Press.

Harnad, S. 1990. The symbol grounding problem. Physica D. Nonlinear Phenomena 42:335–346.

Haugeland, J. 1985. Artificial Intelligence: The Very Idea. Cambridge, MA: MIT Press.

Hurford, J. 2007. Origins of Meaning. Oxford: Oxford University Press.

Hurford, J. 2012. Origins of Grammar. Oxford: Oxford University Press.

288 Vyvyan Evans

Isenberg, N., D. Silbersweig, A. Engelien, S. Emmerich, K. Malavade, B. Beattie, A. C. Leon, and

E. Stern. 1999. Linguistic threat activates the human amygdala. Proceedings of the National
Academy of Sciences of the United States of America 96:10456–10459.

Jackendoff, R. 1983. Semantics and Cognition. Cambridge, MA: MIT Press.

Jackendoff, R. 1987. Consciousness and the Computational Mind. Cambridge, MA: MIT Press.

Johnson, M. 1987. The Body in the Mind. Chicago: University of Chicago Press.

Johnson-Frey, S. H. 2004. The neural bases of complex tool use. Trends in Cognitive Sciences 8 (2):

Kaschak, M., and A. Glenberg. 2000. Constructing meaning: The role of affordances and
grammatical constructions in sentence comprehension. Journal of Memory and Language

Klatzky, R. L., J. W. Pellegrino, B. P. McCloskey, and S. Doherty. 1989. Can you squeeze a tomato?
The role of motor representations in semantic sensibility judgements. Journal of Memory and
Language 28:56–77.

Kranjec, A., and A. Chatterjee. 2010. Are temporal concepts embodied? A challenge for cognitive
neuroscience. Frontiers in Psychology 1 (240): 1–9.

Lakoff, G., and M. Johnson. 1980. Metaphors We Live By. Chicago: University of Chicago Press.

Lakoff, G., and M. Johnson. 1999. Philosophy in the Flesh. New York: Basic Books.

Lakoff, G., and R. Núñez. 2001. Where Mathematics Comes From: How the Embodied Mind Brings
Mathematics into Being. New York: Basic Books.

Langacker, R. W. 1987. Foundations of Cognitive Grammar. Vol. 1. Stanford, CA: Stanford Univer-
sity Press.

Langacker, R. W. 2008. Cognitive Grammar: A Basic Introduction. Oxford: Oxford University Press.

LeDoux, J. E. 1995. Emotion: Clues from the brain. Annual Review of Psychology 46:209–235.

Lindemann, O., P. Stenneken, H. T. van Schie, and H. Bekkering. 2006. Semantic activation
in action planning. Journal of Experimental Psychology: Human Perception and Performance

Machery, E. 2007. Concept empiricism: A methodological critique. Cognition 104:19–46.

Mahon, B. Z., and A. Caramazza. 2008. A critical look at the embodied cognition hypothesis and
a new proposal for grounding conceptual content. Journal of Physiology, Paris 102:59–70.

Martin, A., and L. L. Chao. 2001. Semantic memory and the brain: Structure and processes.
Current Opinion in Neurobiology 11:194–201.

Mithen, S. J. 1996. The Prehistory of the Mind: A Search for the Origins of Art, Religion, and Science.
London: Thames and Hudson.
What’s in a Concept? 289

Moore, K. E. 2006. Space-to-time mappings and temporal concepts. Cognitive Linguistics 17 (2):

Newell, A., and H. A. Simon. 1972. Human Problem Solving. Englewood Cliffs, NJ: Prentice-Hall.

Pinker, S. 1984. Language Learnability and Language Development. Cambridge, MA: Harvard Univer-
sity Press.

Prinz, J. 2002. Furnishing the Mind: Concepts and Their Perceptual Basis. Cambridge, MA: MIT

Pulvermüller, F. 1999. Words in the brain’s language. Behavioral and Brain Sciences 22:253–279.

Pulvermüller, F., O. Hauk, V. V. Nikulin, and R. J. Ilmoniemi. 2005. Functional links between
motor and language systems. European Journal of Neuroscience 21:793–797.

Pylyshyn, Z. W. 1984. Computation and Cognition: Towards a Foundation for Cognitive Science.
Cambridge, MA: MIT Press.

Rizzolatti, G., and L. Craighero. 2004. The mirror-neuron system. Annual Review of Neuroscience

Rosch, E. 1975. Cognitive representation of semantic categories. Journal of Experimental Psychology

104 (3): 192–233.

Rosch, E. 1978. Human categorization. In Advances in Cross-Cultural Psychology, vol. 1, ed.

N. Warren, 1–72. London: Academic Press.

Rosch, E., and C. B. Mervis. 1975. Family resemblances: Studies in the internal structure of
categories. Cognitive Psychology 7 (4): 573–605.

Shapiro, L. 2010. Embodied Cognition. London: Routledge.

Sinha, C., V. da Silva Sinha, J. Zinken, and W. Sampaio. 2011. When time is not space: The social
and linguistic construction of time intervals in an Amazonian culture. Language and Cognition 3
(1): 137–169.

Spivey, M., M. Tyler, D. Richardson, and E. Young. 2000. Eye movements during comprehension
of spoken scene descriptions. Proceedings of the 22nd Annual Conference of the Cognitive Science
Society, 487–492. Mahwah, NJ: Erlbaum.

Stanfield, R. A., and R. A. Zwaan. 2001. The effect of implied orientation derived from verbal
context on picture recognition. Psychological Science 12:153–156.

Talmy, L. 2000. Toward a Cognitive Semantics. 2 vols. Cambridge, MA: MIT Press.

Taylor, L. J., and R. A. Zwaan. 2008. Motor resonance and linguistic focus. Quarterly Journal of
Experimental Psychology 61:896–904.

Taylor, L. J., and R. A. Zwaan. 2009. Action in cognition: The case of language. Language and
Cognition 1 (1): 45–58.
290 Vyvyan Evans

Thierry, G., P. Athanasopoulos, A. Wiggett, B. Dering, and J. R. Kuipers. 2009. Unconscious effects
of language-specific terminology on preattentive color perception. Proceedings of the National
Academy of Sciences of the United States of America 106 (11): 4567–4570.

Thompson-Schill, S. L., K. J. Kurtz, and J. D. E. Gabrieli. 1998. Effects of semantic and associative
relatedness on automatic priming. Journal of Memory and Language 38:440–458.

Turner, M. 2014. The Origin of Ideas. Oxford: Oxford University Press.

Vigliocco, G., L. Meteyard, M. Andrews, and S. Kousta. 2009. Toward a theory of semantic repre-
sentation. Language and Cognition 1 (2): 219–248.

Weiskopf, D. 2010. Embodied cognition and linguistic comprehension. Studies in History and
Philosophy of Science 41:294–304.

Zwaan, R. A. 2004. The immersed experiencer: Toward an embodied theory of language compre-
hension. In The Psychology of Learning and Motivation, ed. B. H. Ross, 35–62. New York: Academic

Zwaan, R. A., and R. H. Yaxley. 2003. Hemispheric differences in semantic-relatedness judgments.

Cognition 87:79–86.
11 Where Are the Concepts? What Words Can and Can’t Reveal

Barbara C. Malt, Silvia P. Gennari, Mutsumi Imai, Eef Ameel, Noburo Saji,
and Asifa Majid

11.1 Introduction

11.1.1 Overview
To study concepts, cognitive scientists need to be able to identify them. The prevail-
ing assumption has been that general-purpose, nonlinguistic concepts are revealed by
words such as triangle, table, and robin. But languages vary dramatically in how they
carve up the world by name.1 Either these concepts are heavily language dependent, or
the words of a language cannot be a direct route to them. In this chapter we argue that
the second of these possibilities is true. We illustrate our point with a study of words
for human locomotion. This study shows that shared conceptual content across four
languages is distinct from the answers suggested by any single language. It supports
the conclusion that words such as triangle, table, and robin do not individuate general-
purpose concepts. However, they can identify underlying components of domain knowl-
edge, which suggests new approaches to understanding conceptual representation.

11.1.2 The Word-Concept Problem

Smith and Medin (1981) opened Categories and Concepts by declaring that concepts give
human experience stability and that mental life would be chaotic without concepts.
Thirty years later, similar sentiments continue to be expressed. Murphy (2002) argues
that concepts hold our mental world together, and Bloom (2004) suggests that a crea-
ture without concepts would be unable to learn. Concepts are considered so fundamen-
tal to human cognition that Fodor (1998) asserted, “the heart of a cognitive science
is its theory of concepts” (vii). If concepts are so important for cognitive scientists to
understand, then cognitive scientists need to be able to identify relevant concepts to
study. In this chapter, we ask what role words can play in identifying them.

1. We use names throughout this discussion to refer to the linguistic labels given to sets of
objects, actions, relations, and properties (not only to those given to individual people, places,
pets, etc.—i.e., proper names).
292 Barbara C. Malt and colleagues

To address this question, we need a working definition of concepts. Murphy (2002)

suggests that concepts are mental representations of classes of objects in the world, and
that they tell their holders what things there are and what properties they have. Carey
(2009) indicates that concepts are units of thoughts that are the constituents of beliefs
and theories. Bloom (2000) suggests that they are mental representations of kinds, and
Solomon, Medin, and Lynch (1999) offer that concepts are building blocks of thought,
and that they have, among their functions, supporting classification, inference, and
conceptual combination. Taken together, such remarks suggest that concepts are stable
units of knowledge in long-term memory that represent meaningful sets of entities
in the world and provide the elements out of which more complex thoughts are con-
structed. Although many authors using the term do not provide an explicit definition,
this description seems to capture the general usage (though a few propose alternatives;
see Barsalou 1987; Smith and Samuelson 1997; Prinz 2002). Throughout most of this
chapter, this dominant approach is what we address. In section 11.3, we revisit this use
of the term and consider whether alternative ways of describing mental representations
of the world may be more useful.
Among those using the general notion of concepts just described, the prevailing
assumption seems to be that many important concepts can be easily identified because
they are revealed by words—in fact, for many researchers, the words of English. Smith
and Medin (1981) used English nouns such as hat, fish, triangle, table, and robin to iden-
tify concepts, as did Rosch (e.g., Rosch and Mervis 1975; Rosch et al. 1976) and Lakoff
(1987). More recent investigations have continued in the same vein (e.g., Murphy 2002;
Fodor 1998, who takes English nouns to reveal the stock of basic concepts that might
be innate; and Carey 2009, who says that the concepts that interest her are roughly
the grain of single lexical items, and that representations of word meanings are para-
digm examples of concepts). Even in formal and computational models, discussions
of the formalisms are illustrated with examples using English nouns (e.g., Kruschke
2005; Rogers and McClelland 2004). And this approach has not been limited to those
studying “concepts” per se. Harnad (2005) declares “kinds” to be the world’s potential
affordances to sensorimotor systems, but he goes on to suggest finding out what those
kinds are by opening a dictionary. Work on conceptual combination has taken nouns
such as chocolate and bee or zebra and fish to indicate what concepts are combined (e.g.,
Hampton 1997, 2012; Wisniewski 1996). In work on neural representation of concepts
and on deficits due to brain injury or disease, “conceptual judgment” tasks often entail
responding to nouns (e.g., Kable et al. 2005; Mahon and Caramazza 2007). In cogni-
tive development research, appeals to English nouns such as horse, cow, boot, and sail
are prevalent in identifying children’s concepts (e.g., Bloom 2000; Gelman 2006; Hills
et al. 2009; Keil 1989; Xu 2005). Kelemen and Carey (2007) talk about the meaning of
the word accordion and the concept of accordion interchangeably, and Gentner (2005)
identifies relational concepts by reference to English nouns (gift, weapon, predator, etc.).
Where Are the Concepts? 293

Even discussions of visual object recognition (e.g., Ullman 1998; Ullman, Vidal-Naquit,
and Sali 2002) and of concept representations in nonlinguistic primates and prelinguis-
tic infants (e.g., Phillips and Santos 2007) often identify the concepts or kinds that are
to be recognized or acquired by means of English nouns.
At an intuitive level, this approach seems reasonable. After all, it makes sense to
think that a person who has a grasp of the words cat and chair has concepts of cats
and chairs. If entities in the world fall into natural groupings according to their shared
properties—as has been argued by a number of researchers (e.g., Anderson 1991; Berlin
1992; Hunn 1977; Rosch et al. 1976; Rogers and McClelland 2004)—then it also makes
sense to take English words to capture those groupings and to be associated with a
coherent mental representation of those groupings. For instance, the noun chair will
capture a set of objects sharing properties such as having a seat, a back, and legs, and
supporting a human in a particular position, and people will have a concept involv-
ing knowledge about this set of things. And, if the preceding points are right, there is
no need for concern about the fact that it happens to be English words that are often
invoked, because other languages should have words that capture essentially the same
groupings and are associated with essentially the same concepts.
But from a different perspective, this approach is startling. Word meanings are
highly selective in what elements of experience they encode. Because of this selectivity,
there are many possible ways to map between words and the world (Wolff and Malt
2010). Crosslinguistic research indicates that languages vary dramatically in how they
carve up the world by name. Substantial crosslinguistic variation has been documented
in domains including color, causality, mental states, number, body parts, containers,
motion, direction, spatial relations, and terms for acts of cutting and breaking and
of carrying and holding (see chapters in Malt and Wolff 2010, for detailed illustra-
tions; also Evans and Levinson 2009; Gentner and Goldin-Meadow 2003; Gumperz
and Levinson 1996; Kay et al. 1997; Majid, Boster, and Bowerman 2008; Majid, Enfield,
and van Staden 2006; Majid and Levinson 2011;Malt and Majid 2013; Saji et al. 2011;
Wierzbicka 2009; among others). As the list indicates, this variation occurs even in con-
crete domains labeled by nouns, where structure in the world might seem most likely
to provide salient groupings that would be captured by the words of any language.
For instance, papers in Majid, Enfield, and van Staden (2006) document diversity in
how languages divide up the human body with body part terms, and Malt, Sloman,
Gennari, Shi, and Wang (1999) found diversity in how languages divide up ordinary
household containers (see also Malt, Sloman, and Gennari 2003). Hand versus arm,
bottle versus jar, or dish versus plate may seem to English speakers to be self-evident
distinctions based on obvious discontinuities in the distribution of properties in the
world, but not all languages observe these same distinctions. Even when structure in
the world does produce shared tendencies in meanings (e.g., the joints for body part
terms; Majid 2010), it underdetermines how any given language will map words onto
294 Barbara C. Malt and colleagues

elements of the world. And this diversity is not only a matter of making fewer versus
more distinctions. Languages often partially or substantially crosscut each other in the
sets of entities they group together (e.g., Bowerman 1996a, 1996b; Malt, Sloman, and
Gennari 2003). For example, whereas English speakers distinguish spatial relations of
containment from support (in versus on) and ignore variations in tightness of fit or
attachment to the surface, Korean speakers label relations of tight containment and
attachment with one word (kkita), contrasting with loose containment (nehta), and
loose support (nohta) (Bowerman 1996a, 1996b).
If the heart of a cognitive science is its theory of concepts, then the field risks seri-
ous, even fatal, defects by overlooking the implications of this diversity. As the just-
cited work establishes, differences among languages are not merely cute examples for
cocktail party conversation (cf. de Boinod 2006, on “extraordinary words from around
the world”). Diversity in how languages carve up the world by name is more of a rule
than an exception across many domains. In light of the documented diversity, there are
three logical possibilities for the nature of the relationship between words and underly-
ing, general-purpose concepts that serve as the elements out of which more complex
thoughts are constructed. These possibilities have clear consequences for theories of
mental representations, and for how researchers would need to look for concepts.
The first possibility is that the words of a language do effectively reveal much of the
stock of basic concepts that a person holds. Importantly, given the pervasive crosslin-
guistic variability in naming patterns, this possibility implies that word learning cre-
ates much of the language user’s nonlinguistic representations of the world (cf. Whorf
1956). Under this scenario, it is not possible to hold that any substantial stock of basic
concepts is shared across speakers of different languages. Concepts revealed by English
words will be true of English speakers, those revealed by Spanish words will be true
of Spanish speakers, and so on, and these language-specific sets will be substantially
different from one another. Models of semantic cognition (e.g., Rogers and McClelland
2004) would need to be taken as models of a particular language group and modified to
give a larger role to names in establishing similarity among representations of entities.
Under this perspective, views of conceptual development would need to grant that the
end result of development is a highly language-specific set of concepts (not just word
meanings), regardless of any inborn universal systems of “core cognition” (Carey 2009)
or universal prelinguistic sensitivities to certain conceptual distinctions (e.g., Casasola
and Cohen 2002; Gentner and Bowerman 2009; Göksun, Hirsh-Pasek, and Golinkoff
The second possibility is that the stock of basic general-purpose concepts is dissoci-
ated to some notable extent from the large differences in naming patterns, and it is
therefore impossible to use words to identify these concepts. After all, much learning
about the world comes from direct interaction, rather than through language. Fur-
thermore, attention to and memory for information in the world may be shaped in
Where Are the Concepts? 295

part by cognitive constraints (such as limits on processing capacity) and evolutionary

influences (such as special sensitivity to survival-relevant information) that are inde-
pendent of the language spoken (e.g., Bock, Carreiras, and Meseguer 2012; Nairne and
Pandeirada 2008; Seligman 1971; Willems et al. 2010). Although speakers of all lan-
guages do not make the same bottle versus jar naming distinction that English speakers
do (Malt et al. 1999), they are likely to note the same differences among containers of
different size and shape regarding their suitability for storing and extracting various
types of substances. This dissociation possibility suggests that nonlinguistic representa-
tions can be substantially shared despite linguistic variability, while still allowing that
language could have some influence on mental representations. Under this possibility,
there may be many widely shared general-purpose representations. There may also
be some nonlinguistic representations that are shaped by the language spoken, but
the shaping will not necessarily fully align the conceptual content with individual
words. Crucially, if linguistic and nonlinguistic representations are distinct and only
loosely linked, then words are no longer direct vehicles to general-purpose concepts.
The words of a language cannot routinely and straightforwardly be used to identify a
person’s concepts.
Empirical evidence supports a dissociation in at least some domains. Malt et al.
(1999) found strong correspondence in similarity judgments for common household
containers among speakers of three languages despite distinctly different naming
patterns, and Ameel, Storms, Malt, and Sloman (2005) found similar results for Belgian
speakers of French versus Dutch (see also Kronenfeld, Armstrong, and Wilmoth 1985).
Also using artifacts, Saalbach and Imai (2007, 2012) found no influence of noun-
classifier categories on Mandarin speakers’ object judgments in several tasks (although
they did find a small influence in two other tasks; see also Huettig et al. 2010).
Similar findings also exist for other domains. Comparing seventeen languages for
color naming versus color sorting, Roberson, Davies, Corbett, and Vandervyver (2005)
found considerable similarity in grouping behavior despite substantial variation in
naming. Munnich, Landau, and Dosher (2001) found that Korean and English speak-
ers’ memory for spatial locations varied less than their naming of the locations (see
also Coventry, Valdés, and Guijarro-Fuentes 2010), and several studies have found dis-
sociations between labeling of actions and attention to or memory for elements of the
actions (e.g., Gennari et al. 2002; Papafragou, Hulbert, and Trueswell 2008; see also
Sauter, LeGuen, and Haun 2011, on similar dissociations for emotional expression).
If the evidence and arguments just cited are valid, this could mean that the second
possibility above is right, the first is wrong, and words must be thrown out as a way of
accessing concepts.
However, a third possibility, less obvious than the first two, also exists and needs to
be evaluated. This third possibility is that the relation of words to the stock of general-
purpose concepts is not as straightforward as current practice has taken it to be (as also
296 Barbara C. Malt and colleagues

suggested in the second possibility), but still, if examined in the right way, words may
reveal something useful about conceptual representations shared across speakers of dif-
ferent languages. By looking beyond individual words from a language as if they pro-
vide a direct route to concepts, and applying more sophisticated techniques to extract
structure from individual language naming or naming aggregated across languages, it
may be possible to discern shared elements of meaning that indicate constraints on
cross-language variability and reflect some common underlying aspects of how knowl-
edge of the domain is represented. In this case, words may retain utility in identifying
a shared understanding of a domain. What they reveal, however, may not necessarily
match the traditional idea of units of knowledge representing meaningful sets of enti-
ties in the world. We consider alternative ways of thinking about conceptual represen-
tation in section 11.3.
In short, linguistic diversity in naming patterns has been well documented across
many domains, but it is rarely considered in the study of concepts. It has potentially
profound implications for this study, but the exact implications have not been pinned
down as of yet. Next, we discuss some data that evaluate different ways of tapping into
the information provided by words about conceptualization of a domain and help dis-
criminate among the three possibilities. These data illustrate what words reveal under
different approaches and what an appropriate use of words is for researchers whose
interest is in underlying nonlinguistic representations rather than in knowledge about
the word meanings of a particular language. In doing so, the data specify some key
implications of linguistic diversity for the study of concepts.

11.1.3 Additional Considerations

Before turning to this illustration, we note two points about past treatment of the rela-
tion of words to concepts in the literature.
First, some, or perhaps many, researchers who use words of English (or another
language) to point to concepts may only be using the words as convenient shorthand.
They do not necessarily have an explicit commitment to a strong Whorfian position
regarding the alignment of words and the stock of basic general-purpose concepts.
However, being noncommittal on the issue is problematic once the pervasive crosslin-
guistic diversity in naming patterns is recognized. Either words must largely determine
these concepts, or they cannot serve as useful shorthand for them. If the second is
true, a separate account is needed of what constitutes these concepts and how they can
be identified. Some researchers may also hold a somewhat more nuanced view of the
word-concept relation. For instance, Murphy (2002), despite using English words to
identify concepts throughout most of his book, argues in a chapter on word meaning
that word meanings do not always have a simple one-to-one relation to concepts, and
that meanings may be composed of a part of a concept or built out of multiple ones.
If taken as his real position, this leaves Murphy the problem of how to identify what
Where Are the Concepts? 297

the more basic general-purpose concepts actually are—an odd dilemma in a book that
is primarily about concepts. Rogers and McClelland (2004) provide their own answer
to what the concepts are. Echoing Rosch and Mervis (1975) and others, they take the
correlational structure of the world to provide the ultimate answer, with mental repre-
sentations of objects clustering based on similarity of properties. Their inferences about
what the similarity relations are, however, come only from considering English nouns.
They assume that the correlational structure of the world produces concepts corre-
sponding to chair, table, and so forth. They do not provide any independent means
of verifying either structure in the world or conceptual structure—again, problematic
without a commitment to the alignment of words and concepts (cf. Majid and Huettig
Second, there is one sense of the term concept in which it must be true that the
words of a language effectively reveal the concepts held by its speakers. That is when
concept is used to mean the knowledge associated with words. In this sense, if English
bottle is applied to a somewhat different set of objects and carries somewhat different
featural implications than Spanish botella (Malt et al. 1999; Malt, Sloman, and Gennari
2003), then English and Spanish speakers have acquired somewhat different concepts.
To the extent that the goal of concept researchers is to study exactly those chunks of
knowledge encoded by the words of a particular language (lexicalized concepts), there
is nothing wrong with using words to identify the knowledge to be studied.
But there are several reasons why studying the knowledge associated with words
is not sufficient for understanding conceptual representation more broadly. First, if
lexicalized concepts are taken as constituting the main set of general-purpose concepts
that cognitive scientists should know about, then this approach functionally makes
elements of language—word meanings—the primary medium of thought, whereas
cognitive scientists generally accept that there is a medium in which thought takes
place that is distinct from any language used for external communication (Fodor
1975). Second, by taking word meanings to reveal the stock of basic concepts used in
more complex thought, it commits cognitive science a priori to a strong version of the
Whorfian hypothesis in which it is language that creates that stock of concepts. Finally,
cognitive scientists who use words to identify concepts most often seem to actually
have a different goal than the goal of understanding the knowledge associated with
words of a specific language. Rosch (Rosch and Mervis 1975; Rosch et al. 1976) explicitly
contrasts the view that segmentation of the world is arbitrary with her own hypothesis
that the world contains inherent structure ready to be recognized by human per-
ceivers. Keil (1989) appeals to “causal homeostasis,” suggesting that concepts (in at
least some domains) are formed in recognition of networks of properties in the world
that are causally connected to one another. More recently, as just noted, Rogers and
McClelland (2004) appeal to structure in the world as the determinant of conceptual
content. Bloom (2004) argues at some length that language is a tool for thought but is
298 Barbara C. Malt and colleagues

not the mechanism that gives rise to the capacity to generate and appreciate ideas in
the first place. If word meanings may be composed of a part of a concept or built out
of multiple ones (Murphy 2002; see also Clark 1983), there must be some more basic
stock of conceptual elements that are the building blocks of word meaning. Further-
more, any researcher interested in the possibility of innate concepts, along with those
who aim to study concepts in nonlinguistic primates or in prelinguistic infants, must
be seeking ones that are not created by learning of words of a language. Much of the
time, then, concepts researchers seem to have a goal of studying representations that
are not created by language, even while using the words of one particular language to
find them.
In sum, our goal is not to argue against the idea that words may gain meaning by
association with conceptual content. Nor are we arguing that the content associated
with a word cannot be called a lexicalized concept. Our concern is with what role
words can play in identifying concepts in the sense of general-purpose representations
that are not inherently part of linguistic knowledge.

11.2 A Case Study: Words for Human Locomotion

We now turn to some data on naming patterns in English, Dutch, Spanish, and Japa-
nese for forms of human locomotion (walking, running, skipping, hopping, and so
on). Humans around the world are capable of the same forms of locomotion regardless
of location or culture, suggesting that there may be substantial shared elements in their
understanding of this domain. At the same time, based on previous data for a more
limited sample of locomotion stimuli (produced on a treadmill that varied in speed
and slope; Malt et al. 2008) we had reason to expect some diversity in naming patterns.
This domain therefore provides a useful case study of the relation of naming patterns
to conceptualization. The stimuli were thirty-six video clips, three to four seconds in
length, depicting upright, bipedal human locomotion of as wide a variety as we could
generate. All actions were demonstrated by an American college student. Figure 11.1
shows sample frames from four clips.
To determine the names each language has for the range of human locomotion
depicted, we asked native speakers of the four languages, mostly undergraduates or
graduate students resident in their home countries, to look at each clip and name it.
Participants viewed the clips embedded in a web page, and for each one, answered the
question “What is the woman doing? She is …,” or its translation in the appropriate
language. More details of the methods, as well as analyses not discussed here, can be
found in Malt et al. (2014). Here, we consider the sets of names produced in each lan-
guage, how names are distributed across actions within each language, and whether
there are shared patterns in this distribution across languages. For each analysis, we
consider the implications for understanding how words may relate to concepts.
Where Are the Concepts? 299

Figure 11.1
Sample frames from clips used in the study, showing upright, bipedal human locomotion.

11.2.1 Name Inventories

We first determined the set of names that were used in each language for these actions.
This analysis is most like the usual approach of concept researchers, except that the
standard approach stops with a single language and does not compare the results with
what other languages would suggest. We asked whether the different languages show
consensus on what these concepts would be.
There was diversity in the surface form people used in their responses. For instance,
for a given clip, in English, many participants might have said that the woman is
walking but a few might have said doing a walk and some even said just walks or a walk.
Some may also have said that the woman is walking slowly, or walking fast with her arms
swinging. We counted as an instance of the same name all surface forms containing the
same root word(s) labeling a manner of movement, and we tallied these for each clip
to determine the frequency of names produced across the participants. We considered
each clip’s “dominant” name to be whatever word was produced most often across the
participants in the language group. Clips for which fewer than 30 percent of respon-
dents agreed on a name were considered “mixed.”
300 Barbara C. Malt and colleagues

Table 11.1
Inventory of locomotion names


English Dutch Spanish Japanese

creep hinkelen caminar aruku

gallop huppelen correr hashiru
hop joggen marchar sukippu-suru
jog lopen saltar ashibumi-suru
jump marcheren trotar kenken-suru
leap rennen koushin-suru
march slenteren janpu-suru
run sluipen
skip springen
stomp stappen
walk wandelen
power walk

Note. Boldface indicates terms that have more than one morpheme.

It is worth considering first what concepts would be identified by simple single-

morpheme words. These words are most like the ones generally considered to pick
out basic-level concepts in the psychological literature (e.g., Murphy 1988; Rosch
and Mervis 1975). The monomorphemic words that emerged as dominant for at least
one film clip in a language are given in table 11.1.2 This tally makes clear that if there
are universally shared concepts of this domain, these words do not directly reveal
what they are. The different languages provide different answers about what that set
would be.
There are also arguments for seeing if conventional names that have more than
one morpheme can help reveal the most basic concepts. Brown (1958) observed that
when people name at the “level of usual utility,” that name is occasionally longer
than a less-used name. For instance, a certain fruit is more likely to be called strawberry
than fruit. A small number of Rosch and colleague’s (1976) “basic level” terms were of

2. There is considerable variation between Netherlands and Belgian Dutch, and also in varieties
of Spanish and English spoken around the globe. Readers may differ in their naming patterns
from those reported here due to such variation. Most notably, lopen has a substantially different
use in the Netherlands than in Belgium, being used primarily for fast locomotion in Belgium and
slow locomotion in the Netherlands.
Where Are the Concepts? 301

this nature (e.g., screwdriver, airplane). Furthermore, across languages, the same notion
may be expressed in different forms: a male child in English is monomorphemic
boy, but in Yucatec, it is xi’pal, containing morphemes for male and for child, and in
Spanish, it is muchacho (or niño), a single root with a masculine suffix (Lucy 2010).
Along those lines, for speakers of Japanese, limiting consideration to simple monomor-
phemic responses risks excluding some common action labels. Japanese speakers use a
construction consisting of an action noun plus the light verb suru (do) to refer to some
actions. Such phrases are similar to English doing a jump or doing a march step, except
that some are fixed, conventional, and common labels for actions. Boldfaced names
in table 11.1 are those added under this version of the tally. Under this count, the
Japanese number of unique names is similar to that for Spanish, but both remain below
the level of English and Dutch. The basic observations from before remain intact. The
languages differ markedly in the inventory of names they display for this domain, and
so names do not identify a universal set of shared concepts.
Could the speakers really vary so much in their concepts of this domain? And could
they really lack concepts of some of the actions, as suggested by the clips designated
mixed, for which there is no agreed-on name? For instance, do Japanese speakers lack
a concept of running in place that the other speakers have? The remaining analyses
help address this question, but for now we note that it seems unlikely that American
and Belgian traditions or current lifestyles lead individuals to develop more highly dif-
ferentiated locomotion concepts, overall, than Argentinean and Japanese experiences
do. If there are differences in the conceptual inventory as large as is suggested by the
naming differences, it seems they must be directly created by the language differences,
in line with the strongest form of a Whorfian perspective.
A glance at the entire phrases participants produced, however, shows that speakers
of all the languages are, in various ways, making many more distinctions among the
actions than those reflected in the conventional names. For instance, Spanish speakers
often used modifiers to discriminate among cases of saltar done on one foot versus two
(hop versus jump for English speakers), and speakers of all the languages used location
modifiers to point out the actions done in place. This observation suggests that people
may have concepts important to the domain that are not captured in a single word.
In the anthropological literature, it has been noted that many languages lack labels
equivalent to English plant and animal, as well as for some groupings within these king-
doms, but the groupings still seem to be recognized as indicated by various nonverbal
behaviors (e.g., Berlin 1992). If a similar situation holds here, then, again, we have to
conclude that words are a poor indicator of where the concepts of interest lie (even if
modest cultural differences do exist among the groups).
This first analysis points out the fundamental problem of taking individual words
of any one language to directly reveal concepts: different languages will give different
answers to what the concepts are, and these linguistic differences are likely to exceed
302 Barbara C. Malt and colleagues

the extent of any conceptual differences. We now ask whether other ways of examining
the naming data do a better job at uncovering shared concepts. To the extent that they
do, the results will underscore the fallacy of treating the individual words as if they
directly reveal concepts. At the same time, they will suggest that naming data, exam-
ined in an appropriate way, may still provide some evidence of a shared understanding
of the domain.

11.2.2 Treating Individual Languages’ Naming as Similarity Data

In this analysis, we asked whether commonalities emerge from the naming data of the
four languages if we make use of the full set of names produced by all participants for
all stimuli. Many clips did not produce 100 percent name consensus within a language
group, so even if one language has different dominant names for two clips (e.g., walk
and stroll), and another gives them the same name (e.g., caminar), some speakers of
the first language may have produced the same name for both (e.g., walk), pointing to
a perceived similarity between them. We created name similarity matrices that reflect
the extent to which each pair of objects received the same name across the speakers
of a language. To do this, we assigned, for each participant, a 0 or a 1 to each possible
pair of clips according to whether the person gave the two clips a different name or
the same name (again, tallied according to the guidelines described above). There are
630 such pairs, given the thirty-six clips. We then constructed a similarity matrix for
each language group by summing the distance values for each of the 630 pairs of clips
across the participants in that language group. This use of the data is similar to using
confusion matrices as similarity data (e.g., Rothkopf 1957; Shepard and Chang 1963).
By taking into account the full naming array for each clip and applying scaling tech-
niques to the similarity matrices, commonalities among languages may become more
We correlated the name similarity matrices for each pair of languages to give an
overall sense of the correspondence in the naming patterns, using the Mantel test for
correlation of matrices (Mantel 1967). These correlations are all significant, falling
between 0.65 and 0.82. The correlations indicate that the full patterns of name use,
while diverse, still share substantial commonalities.
We then carried out multidimensional scaling on the matrix for each language (pro-
cedure MDS; SAS Institute 1999). To help interpret the results, additive tree clusters
were drawn on the solutions (Corter 1982; Sattath and Tversky 1977). Solutions for the
four languages are shown in figures 11.2–11.5, with only the top two levels of clusters
for ease of viewing. Labels in the solution refer to the clip names bestowed by the
experimenters (not the name given by participants) so that the solutions can be com-
pared with regard to which clips are grouped together across the languages.
These solutions, while showing similarities, also highlight important differences.
For all four languages, the horizontal dimension is interpretable in terms of the basic


Where Are the Concepts?


Dimension 2



–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5 3
Dimension 1

Figure 11.2
MDS solution based on American English naming data. Clip names refer to names bestowed by the experimenters, not names
produced by participants.



Dimension 2




–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5 3
Dimension 1

Figure 11.3
MDS solution based on Belgian Dutch naming data. Clip names refer to names bestowed by the experimenters, not names produced
by participants.
Barbara C. Malt and colleagues


Where Are the Concepts?


Dimension 2




–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5 3
Dimension 1

Figure 11.4
MDS solution based on Argentinean Spanish naming data. Clip names refer to names bestowed by the experimenters, not names
produced by participants.



Dimension 2




–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5 3
Dimension 1

Figure 11.5
MDS solution based on Japanese naming data. Clip names refer to names bestowed by the experimenters, not names produced by
Barbara C. Malt and colleagues
Where Are the Concepts? 307

biomechanical distinction between an elastic impact-and-recoil motion (characteristic

of running gaits and others such as hopping and jumping) and a pendulum motion
(characteristic of walking gaits), where one foot is on the ground at all times (e.g.,
Alexander 2002; Bennett 1992). (The pendulum-based gaits appear toward the left-hand
side of each solution and the impact-and-recoil toward the right.) However, the exact
spatial layout of clips varies considerably across the languages. The vertical dimen-
sion for the most part seems to reflect a dimension of speed and aggressiveness (with
slower, less aggressive actions lower and faster, more aggressive ones higher for English
and Spanish, and vice versa for Dutch), but the Japanese solution less clearly conforms
to this possibility. The clusters resulting from the additive tree analysis reinforce the
idea that the biomechanical distinction is salient for two of the languages—Dutch and
Japanese—which have top-level clusters separating essentially the same sets of clips
(though Dutch places the RUN_FAST clip just into the cluster of pendulum motions).
English and Spanish clusters are less like the Dutch and Japanese results: for English,
running actions cluster with pendulum motions at the top level, and for Spanish, walk-
ing backward and several forms of marching combine with impact-and-recoil motions,
as well as walking in place. In the next level of clusters, within these top-level clusters,
each language more or less separates the faster, more aggressive pendulum actions from
slower, more cautious pendulum actions, but the exact composition of the clusters is
variable. The Dutch solution, in particular, does not honor this separation as much as
the others.
These solutions indicate that the naming patterns of the four languages reflect a
shared orientation to the same dimensions of the movements. This outcome supports
the idea that speakers of the four languages may have more in common in their percep-
tion of the domain than their name inventories indicate. In light of the variability of
the additive tree clusters across the four solutions, though, it remains difficult to say
exactly what could be identified as shared discrete concepts in the traditional sense.

11.2.3 Treating Aggregated Naming as Similarity Data

Last, we created a multidimensional scaling solution combining the naming data of
participants in all four language groups. MDS by its nature looks for commonalities
in the data and, to the extent that it finds them, produces a coherent solution. If a
coherent solution emerges, this result would again support the idea of a shared concep-
tualization of the domain while underscoring the inadequacy of individual words of a
single language to reveal it. Compatible with this possibility, Regier, Kay, and Khetarpal
(2007) examined the highly variable color terms across 110 languages and found
evidence for a general well-formedness constraint of maximizing similarity within
lexical categories and minimizing it across categories. Khetarpal, Majid, and Regier
(2009) obtained similar results for spatial terms of nine languages. More similar to the
current analysis, Majid, Boster, and Bowerman (2008) aggregated across twenty-eight
308 Barbara C. Malt and colleagues

languages’ verbs for cutting and breaking actions and found a shared set of dimensions
underlying the different naming patterns. With only four languages to combine here,
our data provide less to aggregate, but the smaller sample of languages is more like that
typically available to psychologists working on concepts. From that perspective, it is
particularly useful to see if a coherent conceptual space emerges from the data of only
four languages.
We carried out multidimensional scaling as before. Figure 11.6 shows the two-
dimensional solution, again with additive tree clusters imposed on it. This solution
shows a neat horizontal separation of the impact-and-recoil motions (toward the
right) from the pendulum motions (toward the left). The vertical dimension again
appears to reflect something like speed and aggressiveness of the actions, with slower,
less aggressive actions toward the top and faster, more aggressive ones toward the
At the top level of clustering, the impact-and-recoil motions are separated from the
pendulum-based ones, with the exception of the TROT clip falling into the pendu-
lum cluster. The placement of the TROT clip is most likely because of the qualities of
the particular action implemented in the clip, which was a bouncy motion but with
little or no evidence of both feet being off the ground at the same time. Within these
clusters, subclusters separate the running clips from the other impact-and-recoil actions,
and separate the true pendulum motion clips from the intermediate TROT clip. These
clusters are thus readily interpretable, although they do not seem to map directly onto
the words of any of the languages.
So, the naming data when aggregated across the four languages provide more
indication of a systematic conceptual space than looking at scaling solutions of the
four languages individually does. This shared space emerges out of the noisiness of
the individual name inventories, which make different distinctions and numbers of
distinctions. Because MDS can discover commonalities in data, but cannot invent them,
the simplicity of the solution is evidence in favor of a shared underlying understand-
ing of the domain. At the same time, though, if the identified additive tree clusters are
taken to indicate discrete concepts within this space, they do not seem to be picked out
by words of the languages. This outcome again suggests a shared conceptualization of
the domain that is not revealed by the words of any single language.

11.2.4 Conclusions from Using Names to Reveal Conceptual Space

These analyses demonstrate that different languages would tell us different things
about concepts if we were only to look at their name inventories. If there are shared
general-purpose concepts, then the words of any one language do not directly reveal
what they are. Moreover, different ways of counting names (monomorphemic only
versus including multimorphemic ones) produce somewhat different answers. Even if
we were to adopt a strong version of the Whorfian hypothesis and assume that words


Where Are the Concepts?


Dimension 2




–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5 3
Dimension 1

Figure 11.6
MDS solution based on the four languages’ combined naming data. Clip names refer to names bestowed by the experimenters, not
names produced by participants. Adapted from Malt, Ameel, Imai, Gennari, Saji, and Majid (2014).
310 Barbara C. Malt and colleagues

of a language do directly reveal basic concepts for speakers of that language, it is not
clear which way of counting would be the right one to use, and using names as direct
pointers to concepts would remain problematic.
Despite the diversity in the name inventories, other ways of analyzing the data show
commonalities underlying the naming patterns. In particular, scaling of the combined
naming data of four languages produces a coherent and interpretable solution, suggest-
ing a shared orientation to certain dimensions of the space. Still, the clusters within
the scaling solution do not neatly correspond to those labeled by the names of the
languages, raising questions about what, if anything, can be identified as discrete con-
cepts in the traditional sense.

11.3 Issues and Implications

11.3.1 Implications for the Status Quo

The data make a strong case that the relation of words to concepts is not straightfor-
ward. Pervasive linguistic diversity, amply documented in other research but rarely
taken into consideration within “concepts” research, is by itself cause for serious
concern. It would still be possible to preserve a commitment to word-concept align-
ment by subscribing to a strong version of the Whorfian idea that language shapes
thought. The aggregated naming data, however, suggest that any conceptual differ-
ences are less than implied by differences in word inventories. This evidence is also
consistent with past findings in other domains and the observation that attention to
and understanding of aspects of the world are likely shaped by multiple forces that may
include but are not limited to language (see section 11.1.2). It seems unavoidable to
conclude that researchers need to stop relying on the word inventories of English, or
any other single language, to know what constitutes the concepts of a domain.
In light of our data and arguments, what should be made of the existing literature on
concepts? We suggest that what is being studied in many cases—those where probing
of knowledge comes in the form of asking for responses of various sorts to the words of
English or another language—is people’s knowledge associated with the words. There
is nothing wrong with investigating knowledge associated with words, and making this
clarification does not cast doubt on the value of the work. It does, however, suggest the
need to reframe the understanding of what the work reveals, because this methodology
does not directly shed light on general-purpose nonlinguistic concepts, possibly shared
across speakers of different languages. The only way for researchers to argue that their
work directly investigates nonlinguistic concepts is by making an explicit commitment
to a strong Whorfian perspective. Whichever approach the researcher takes, it should
always take into account the existing pervasive crosslinguistic diversity and the fact
that word inventories can only be used as the means of tapping units of knowledge by
committing to the notion that the units accessed are language specific.
Where Are the Concepts? 311

It might be possible to argue for a direct alignment of words with general-purpose

concepts for some domains on the grounds that in certain domains, word meanings
should be broadly shared across speakers of different languages. Such word meanings
could be hypothesized to exist for any innate concepts, and for concepts that are devel-
oped in a similar way across cultures due to cross-cultural recognition of structure in
the world. A logistical complication still arises in that the existence of broadly shared
word meanings cannot be assumed; it would have to be verified by careful crosslinguis-
tic comparison for a domain of interest. An empirical complication also arises in that
in many such domains where crosslinguistic data have been examined, word meanings
have turned out to be more variable than previously imagined. Even though substantial
elements of the human experience of sensory input, emotion, or space, for instance,
might be universal (e.g., Clark 1973; Ekman 1992; Henning 1916), recent investiga-
tions have shown that sensory vocabularies (Majid and Levinson 2011), emotion terms
(e.g., Wierzbicka 1999, 2009) and spatial term inventories and meanings (Bowerman
1996a, 1996b) are considerably variable across languages. Perception of the body or of
human locomotion might be heavily constrained by structure in the stimulus domain
(body joints or biomechanics of movement), but the current work and our earlier work
(Majid, Enfield, and van Staden 2006; Malt et al. 2008) have shown crosslinguistic
diversity in these domains, too. The body part and locomotion work does show that
alongside diversity, structural constraints are reflected in shared elements of meaning.
However, it seems impossible to discern from only a single language what the shared
elements will be and which parts of the patterns are idiosyncratic to the language.
The domain of natural kinds is another for which it has been argued that structure
in the world will give rise to shared elements of meaning (Berlin 1992; Hunn 1977)—in
this case, the structure being clusters of correlated properties separating scientific genera
such as bovines versus equines. Berlin (1992), Hunn (1977), and others (see Malt 1995
for review) document that nonindustrialized cultures tend to recognize by name much
the same sorts of groupings that modern science and more technologically advanced
cultures do. But again, the broad similarities are accompanied by differences. Primary
names that label rich, important, frequently used conceptual content for members
of traditional societies sometimes point to highly impoverished representations for
members of urban societies (such as elm versus oak versus chestnut; Dougherty 1978;
Wolff, Medin, and Pankratz 1999) due to attrition of knowledge bases over cultural
time. Conversely, some significant groupings may lack labels in various languages, as
in the case of labels for plants versus animals mentioned earlier, as well as groupings
within them that are perceived but of lesser cultural utility (see Malt 1995 for review).
Furthermore, although many groupings labeled with primary names may reflect struc-
ture in the world, some may depend on properties such as domesticity or use. In fact,
many English names for familiar naturally occurring entities pick out groupings based
on their role in human lives rather than biology, for example, weed, tree, vegetable, dog
312 Barbara C. Malt and colleagues

(see Malt 1991) and so do not reflect world structure of the sort argued to produce
crosslinguistic similarities. Most likely, the degree to which words of a single language
can directly reveal shared concepts for a domain for speakers of different languages will
fall on a continuum from poor to better (but still imperfect). The cases just discussed
indicate that it may not be possible to discern what is poor and what is better without
detailed empirical study.
Despite the complex nature of the relationship between language and general-
purpose concepts, our data suggest that it is still possible to use linguistic data to gain
insight into something more fundamental about the nature of conceptual space. That
is, combining naming information across languages does seem to provide useful infor-
mation, because the aggregate allows commonalities to emerge over the “noise” of
individual language idiosyncrasies. Sometimes naming data are more readily available
for researchers than data such as similarity judgments. This is particularly true for those
already in a position to collect data from members of different language communities
through simple paper and pencil tasks, or who have access to archived data on naming
for domains of interest. However, it is clear that there are also methodological chal-
lenges involved in collecting naming data across languages, in particular, gaining access
to participants who speak diverse languages and having available the language exper-
tise to administer tasks and interpret data from the various languages. For researchers
whose usual methodologies entail only members of one language group, obtaining
such crosslinguistic data may not be feasible. In those cases, developing other methods
of avoiding the problem of faulty reliance on a word-concept equivalence will be
crucial, such as finding ways to implement measures less dependent on language.

11.3.2 If There Is a Shared Conceptual Space, Why Don’t Individual Naming

Patterns Align Better?
The possibility of a significant dissociation of naming patterns from how people under-
stand a domain nonlinguistically raises the question of why languages diverge in their
naming patterns in the first place. If there are shared experiences across cultures in at
least some domains, why wouldn’t the naming patterns for those domains align more
closely? The answer most likely lies in the facts that (a) words encode only a frac-
tion of the richness of experience, and (b) the meanings encoded in any particular set
are shaped in part by forces independent of the conceptual representations of current
speakers. Some elements of arbitrariness in the early development of a set of meanings
will exist, simply because of the limitation on how much words can encode. As the
language continues to evolve, external forces such as contact with other languages,
waxing or waning of the importance of a domain to the culture, the introduction of
new entities within the domain that need to be named and the order in which they
enter the culture, and influences from the syntax and morphology of language on how
distinctions can be expressed can alter the set of words available in a domain and the
Where Are the Concepts? 313

meanings associated with each word (e.g., Aitchinson 2001; see Malt, Gennari, and
Imai 2010 for discussion). Because each language evolves on its own path, and the
current state of a language is always a function, in part, of past states (e.g., Dunn et al.
2011), languages will develop their own patterns of naming that may be only loosely
related to one another, as well as only loosely related to current ways of thinking about
the world by their speakers.

11.3.3 Implications for Property Projection

Promoting inferences has been taken to be a critical benefit of putting names on things
(e.g., Gelman and Markman 1986; Heit 2000; Osherson et al. 1990; Quine 1970). The
partially arbitrary nature of linguistic categories highlights the fact that inferences
will not always be well circumscribed by named categories. For instance, consider the
English names chair, knife, and spoon. A beach chair is light, but an armchair is heavy; a
desk chair has a rigid seat, but a beanbag chair does not; a kitchen chair may be wood
and a folding chair metal. A kitchen knife is intended for cutting but a frosting knife
for spreading; a soup spoon is for scooping up liquids but a slotted spoon for draining
them; a hair brush is for smoothing but a scrub brush is for scrubbing. This imperfect
matching of properties to names is not limited to artifacts: although Chihuahuas and
Siberian Huskies share the property of domesticity that projects across things called
dog, other properties such as size, strength, home (small or large, indoors or out), and
ability to withstand cold may be more shared with things called cat or wolf. Relying
heavily on names to make inferences, then, might often produce results that would
turn out to be inaccurate. To the extent that inferences about unseen properties are
accurate, they may be made by projecting on the basis of observed commonalities
with familiar instances more often than is generally acknowledged (see also Gelman
and Davidson 1990). Despite Rogers and McClelland’s (2004) problematic use of
English words to identify concepts, their model provides a useful demonstration of
how such inferences can be drawn without an explicit process of categorization inter-
vening between perception of an entity and property projection.

11.3.4 Implications for the Notion of Concepts

The combined naming data from the four languages produces a fairly interpretable
scaling solution suggesting that naming is driven by some fundamentally shared under-
standing of the characteristics of human locomotion. Different languages may make
different numbers of distinctions and vary somewhat in what kinds of distinctions
are being made, but they seem to be drawing on a shared pool of dimensions. Similar
observations have been made for other domains (e.g., Jones 2010; Majid, Boster, and
Bowerman 2008; see also Slobin 2010 on locomotion). Yet the scaling solutions, while
suggesting some shared understanding of the domain across speakers of different
languages, and some salient groupings of instances, still leave it unclear exactly what
314 Barbara C. Malt and colleagues

units of knowledge should count as the most fundamental, basic concepts of the
domain. Different levels of the cluster analysis